CN115713844B - Alarm method and system - Google Patents
Alarm method and system Download PDFInfo
- Publication number
- CN115713844B CN115713844B CN202211269854.0A CN202211269854A CN115713844B CN 115713844 B CN115713844 B CN 115713844B CN 202211269854 A CN202211269854 A CN 202211269854A CN 115713844 B CN115713844 B CN 115713844B
- Authority
- CN
- China
- Prior art keywords
- chaotic
- ultrasonic
- alarm
- signals
- relu activation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Measurement Of Velocity Or Position Using Acoustic Or Ultrasonic Waves (AREA)
Abstract
The application relates to the technical field of industrial manufacturing and discloses an alarming method and system, wherein the method is applied to the alignment and assembly process of large and ultra-large components or accessory components, and comprises the steps of acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of the components; preprocessing the acquired chaotic ultrasonic signals; inputting the preprocessed chaotic ultrasonic signals into a preset neural network, and outputting the position information of key nodes of the component; judging whether to alarm or not based on the position information and the set threshold value; the method solves the problem that the prior alarm system alarms the abnormal state of the component in the alignment and assembly process, and has the effect of improving the alarm accuracy of the alarm system on the abnormal state of the component in the alignment and assembly process.
Description
Technical Field
The present disclosure relates to the field of industrial manufacturing technologies, and in particular, to an alarm method and system.
Background
Alignment and assembly of large or ultra-large components such as aircraft parts, rocket bodies, and the like is a problem in the field of industrial manufacturing that is difficult. If the components or installed equipment deviate during alignment and installation, the quality of the final product is affected and even economic losses and casualties are incurred. During the alignment and assembly process, warning of possible abnormal states or movements of the component or its accessory assembly components, for example controlling the mechanical arm of the industrial robot on which the component or assembly part is mounted, the movement speed of which is too slow or suddenly stopped; the combination of the component or the assembly part and the mechanical arm is not firm, and the posture of the component or the assembly part is deviated; the falling off of the member or the assembly part causes the condition that the preset boundary is exceeded, etc. The risk occurrence probability can be effectively reduced, and the stability of the whole alignment or installation system is improved.
In an assembly site, the existing alarm system is affected by multipath interference and noise, has lower positioning precision on key nodes on a large or ultra-large component or an accessory assembly component thereof, and is easy to cause the condition of no alarm or false alarm; and direct transmission of one path of signals is easy to be blocked in the moving process of the mechanical arm or the hoisting mechanism, so that error warning can occur due to deviation when the position information of the key node of the component is solved through a positioning algorithm.
Aiming at the related technology, the inventor finds that the prior alarm system has lower positioning precision and is easy to generate false alarm and non-alarm because the prior alarm system is influenced by multipath interference and noise on a large or ultra-large component or a key node on an accessory assembly component thereof in the industrial production process, and the prior alarm system has larger deviation on the positioning of the key node through an algorithm when one path of direct wave signal in the positioning process in a site is blocked, so that the false alarm condition occurs.
Disclosure of Invention
In order to improve the alarm accuracy of an alarm system on abnormal states of the components or the assembly parts, and accurately alarm the alignment and assembly processes of the components or the assembly parts, and reduce the situations of false alarm and non-alarm, the application provides an alarm method and an alarm system.
In a first aspect, the present application provides an alert method.
The application is realized by the following technical scheme:
an alarm method for alignment and assembly of large and ultra-large components or accessories, comprising the steps of,
acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of a component;
preprocessing the acquired chaotic ultrasonic signals;
inputting the preprocessed chaotic ultrasonic signals into a preset neural network, and outputting the position information of key nodes of a component;
and judging whether to alarm or not based on the position information and the set threshold value.
The present application may be further configured in a preferred example to: the step of judging whether the alarm is given according to the set threshold value by the position information comprises,
acquiring position information of key nodes on a component;
and judging whether the position boundary of the key node exceeds a preset boundary range and whether the movement speed exceeds a threshold value according to the position information of the key node, and alarming if the position boundary exceeds the threshold value.
The present application may be further configured in a preferred example to: the neural network comprises a backbone network, a coarse positioning network and a fine positioning network;
the output signals of the backbone network are input into the coarse positioning network and the fine positioning network;
The output signal of the coarse positioning network is input to the fine positioning network.
The present application may be further configured in a preferred example to: the backbone network consists of five bottleneck structures, a global average pooling layer and two attention modules;
wherein, the convolution layer of the first bottleneck structure is two-dimensional convolution, and the convolution layers of the rest bottleneck structures are one-dimensional convolution;
the output signal of the first bottleneck structure is input into the global average pooling layer and then sequentially input into the rest bottleneck structures;
the output signal of the last bottleneck structure is sequentially input to the attention module.
The present application may be further configured in a preferred example to: the bottleneck structure comprises 3 convolution layers, 3 ReLU activation functions and 1 downsampling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated output signals into the downsampling layer.
The present application may be further configured in a preferred example to: the coarse positioning network comprises 3 convolution layers, 3 ReLU activation functions and 1 adaptive average pooling layer;
The convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated signals into the adaptive average pooling layer.
The present application may be further configured in a preferred example to: the fine positioning network comprises 5 convolution layers, 5 ReLU activation functions and 1 adaptive average pooling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated signals into the adaptive average pooling layer.
The present application may be further configured in a preferred example to: the step of inputting the preprocessed chaotic ultrasonic signals into a preset neural network and outputting the position information of the key nodes of the component further comprises the steps of,
the neural network is trained using a sample dataset, wherein the sample dataset is divided into a simulated sample set and a real sample set.
The chaotic ultrasonic signals which are generated according to simulation software and are preprocessed are used as simulation samples in the simulation sample set, and the simulated position information of the chaotic ultrasonic signals is used as a label of the simulation samples;
Taking the chaotic ultrasonic signal transmitted by the preprocessed preset ultrasonic generating device as an actual sampling sample in the actual sampling set, and taking the position information of the preset ultrasonic generating device as a label of the actual sampling sample;
dividing the simulation sample set and the actual sampling sample set into a simulation training set, a simulation test set, an actual sampling training set and an actual sampling test set according to a preset proportion;
inputting the simulation training set and the corresponding label into the neural network for training, and inputting the simulation testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a first threshold;
and inputting the actual acquisition training set and the corresponding label into the trained neural network for training, inputting the actual acquisition testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a second threshold value, and outputting the neural network.
The present application may be further configured in a preferred example to: the preprocessing the acquired chaotic ultrasonic signal includes,
supplementing 0 to the pre-generated chaotic ultrasonic signal to enable the length of the pre-generated chaotic ultrasonic signal to be the same as that of the acquired chaotic ultrasonic signal;
And splicing the chaotic ultrasonic signal subjected to 0 supplementation with the acquired chaotic ultrasonic signal with the same frequency band to obtain a preprocessed chaotic ultrasonic signal.
In a second aspect, the present application provides an alert system.
The application is realized by the following technical scheme:
an alarm system comprises an ultrasonic generation module, an alarm ultrasonic acquisition device group, an alarm data processing module, an alarm judging module and an alarm module;
the ultrasonic generation modules are respectively arranged on key nodes of the component and are used for transmitting chaotic ultrasonic signals in different frequency bands;
the alarming ultrasonic acquisition device group is used for acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of the component;
the alarm data processing module is connected with the alarm ultrasonic acquisition device group and is used for preprocessing the acquired chaotic ultrasonic signals;
the alarm judging module is used for outputting the position information of the key nodes of the component through a preset neural network and judging whether an alarm is given or not based on the position information and a set threshold value;
and the alarm module is used for automatically alarming based on the judging result of the alarm judging module.
In a third aspect, the present application provides a computer device.
The application is realized by the following technical scheme:
a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of any one of the alert methods described above when the computer program is executed.
In a fourth aspect, the present application provides a computer-readable storage medium.
The application is realized by the following technical scheme:
a computer readable storage medium storing a computer program which when executed by a processor performs the steps of any of the alert methods described above.
To sum up, compared with the prior art, the beneficial effects brought by the technical scheme provided by the application at least include:
an alarming method applied to the in-field component alignment and assembly process of large and ultra-large components is characterized in that chaotic ultrasonic signals of different frequency bands sent out on key nodes of the components are obtained, the obtained chaotic ultrasonic signals are preprocessed, the preprocessed chaotic ultrasonic signals are input into a preset neural network, position information of the key nodes of the components or assembly components is output, and whether the alarming is carried out is judged according to a set threshold value based on the position information. The chaotic ultrasonic signals in different frequency bands are not interfered with each other, so that crosstalk between the chaotic ultrasonic signals is reduced, the distance measurement information contained in the preprocessing spliced signals is ensured to come from the same ultrasonic generating device, and the neural network is beneficial to accurately outputting the position information of the key node; the situation that errors are generated due to phase matching caused by the periodicity of signals, and positioning accuracy is affected is improved. The possibility of false alarm and non-alarm of an alarm system is reduced; the boundary and the threshold value of the alarm can be judged according to different adjustment of the task types, and a new alarm system is not required to be reset for the task installed or docked in the same site; compared with the traditional positioning algorithm, the neural network output key node position information has better anti-interference capability, and can output more accurate position information under the condition that the input signal contains a small amount of interference, so that the probability of false alarm and non-alarm occurrence of an alarm system is effectively reduced; the situation that the direct path of signals is blocked by an actuating mechanism or an assembly part in the field to cause false alarm in the traditional alarm method process is improved.
Drawings
Fig. 1 is a flow chart of an alarm method according to an exemplary embodiment of the present application.
Fig. 2 is a main block diagram of a neural network of an alarm method according to an exemplary embodiment of the present application.
Fig. 3 is a main network structure diagram of a neural network according to an exemplary embodiment of the present application.
Fig. 4 is a schematic diagram of a bottleneck structure of a backbone network of a neural network according to an exemplary embodiment of the present application.
Fig. 5 is a diagram of a coarse positioning network structure of a neural network according to an exemplary embodiment of the present application.
Fig. 6 is a diagram of a fine positioning network structure of a neural network according to an exemplary embodiment of the present application.
Fig. 7 is a block diagram of an alarm system according to an exemplary embodiment of the present application.
Fig. 8 is a signal transmission schematic diagram of an alarm system according to an exemplary embodiment of the present application.
Detailed Description
The present embodiment is merely illustrative of the present application and is not intended to be limiting, and those skilled in the art, after having read the present specification, may make modifications to the present embodiment without creative contribution as required, but is protected by patent laws within the scope of the claims of the present application.
For the purposes of making the objects, technical solutions and advantages of the embodiments of the present application more clear, the technical solutions of the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by one of ordinary skill in the art based on the embodiments herein without making any inventive effort, are intended to be within the scope of the present application.
In addition, the term "and/or" herein is merely an association relationship describing an association object, and means that three relationships may exist, for example, a and/or B may mean: a exists alone, A and B exist together, and B exists alone. In this context, unless otherwise specified, the term "/" generally indicates that the associated object is an "or" relationship. Embodiments of the present application are described in further detail below with reference to the drawings attached hereto.
At present, the alarm system mainly comprises an alarm system based on manual observation, an alarm system based on electromagnetic waves and an alarm system based on traditional ultrasonic positioning.
Based on the alarm of manual observation, the abnormal conditions in the alignment and assembly process of the large-scale components are alarmed in a human eye observation mode. The alarm mode relies on subjective perception and experience accumulation of people, judgment of abnormal states of components or assembly parts requires monitoring staff to accurately grasp movement conditions, boundary conditions and set thresholds of the components or assembly parts, the observation precision of boundary parameters of the components or assembly parts is not high, and human eyes observe conditions such as incomplete observation and artificial fatigue, so that the accuracy of an alarm system is finally affected.
The electromagnetic wave-based warning system adopts an electromagnetic wave ranging technology, and a radar range finder is placed on the surface of a member to be butted in the aligning and assembling process, so that the distance measurement can be automatically carried out through electromagnetic wave self-receiving and the warning can be carried out according to the distance measurement result. The sensitivity of the alarm system is high, but at the scale of the factory site, the propagation speed of electromagnetic wave is about 3×10 8 m/s, if a tiny disturbance is introduced in the measurement of the system time, the measurement based on electromagnetic waves can generate a great deviation, and the alarm system can generate more false alarms. And because the radar range finder has stronger directivity, when an actuating mechanism or an assembly part appears in the range finding direction, a signal sent by the radar range finder can generate false alarm due to shielding. Furthermore, the positioning algorithms such as traditional TDOA and MUSIC are adopted, a plurality of electromagnetic wave receiving devices are utilized to calculate the position information of a plurality of electromagnetic wave generating devices arranged at key nodes of the component through the positioning algorithm, and the position information is utilizedThe alarm is given, but because of the periodical influence in the signal, the time measurement introduces tiny disturbance, so that the position of the key node is greatly deviated, and the situation of false alarm caused by exceeding a preset boundary is positioned. Meanwhile, the algorithm positioning solving process is easy to be affected by noise or interference of multipath effect, positioning accuracy is affected due to the fact that the positioning modes such as wifi, bluetooth and UWB are not high in positioning accuracy in an assembly or butt joint field due to the characteristic of high propagation speed of superimposed electromagnetic waves, a large number of extra experiments are needed to compensate and correct an algorithm model, design difficulty is high, implementation difficulty and implementation cost are high, and accordingly accuracy of alarming is affected. In addition, when an actuating mechanism or an assembly part appears in the direct transmission direction of electromagnetic waves, electromagnetic wave signals are blocked, so that errors are generated in electromagnetic wave ranging, and the situation that error warning occurs due to deviation of position information of key nodes of a component is solved through an algorithm.
Based on the warning system of traditional ultrasonic positioning, in the aligning and assembling process, place ultrasonic ranging appearance on the surface of the member to be docked, utilize ultrasonic wave to receive spontaneously and carry out ranging and report an emergency and ask for help or increased vigilance according to the ranging result, but like electromagnetic wave warning system, because ultrasonic wave has stronger directionality, when actuating mechanism or equipment part appear in its range finding direction, the signal that ultrasonic ranging appearance sent can produce the mistake and report an emergency and ask for help or increased vigilance because of sheltering from. Or, using a plurality of ultrasonic wave receiving devices, calculating the position information of a plurality of ultrasonic wave generating devices arranged on the key nodes of the component through a positioning algorithm, and alarming by using the position information. However, the ultrasonic signal of the ultrasonic wave is easily affected by the internal periodicity, and the time measurement introduces tiny disturbance due to the periodicity, so that a certain amount of deviation can occur at the position of the final positioning point, and further a false alarm condition is generated. In addition, the algorithm positioning solving process is easily affected by noise or interference of multipath effects, so that positioning accuracy is affected, such as deviation exists between TDOA, MUSIC and the like in application and actual conditions, a large number of additional experiments are needed to compensate and correct an algorithm model, design difficulty is high, implementation difficulty and cost are high, and alarm accuracy is further affected. In addition, when an actuating mechanism or an assembly part appears in the direct transmission direction of ultrasonic waves, signals are blocked, so that errors are generated in ultrasonic ranging, and the situation that error warning occurs when position information of key nodes of a component is deviated is solved through an algorithm.
Referring to fig. 1, an embodiment of the present application provides an alarm method applied to an alignment and assembly process of large and ultra-large components or accessory components, and main steps of the alarm method are described as follows.
S1, acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of a component;
s2, preprocessing the acquired chaotic ultrasonic signals;
s3, inputting the preprocessed chaotic ultrasonic signals into a preset neural network, and outputting position information of key nodes of a component;
and S4, judging whether to alarm or not based on the position information and the set threshold value.
Further, the step of judging whether the alarm is given by the position information and the set threshold value includes,
acquiring position information of key nodes on a component;
and judging whether the position boundary of the key node exceeds a preset boundary range and whether the movement speed exceeds a threshold value according to the position information of the key node, and alarming if the position boundary exceeds the threshold value.
Further, the neural network comprises a backbone network, a coarse positioning network and a fine positioning network;
the output signals of the backbone network are input into the coarse positioning network and the fine positioning network;
the output signal of the coarse positioning network is input to the fine positioning network.
Further, the backbone network consists of five bottleneck structures, a global average pooling layer and two attention modules;
wherein, the convolution layer of the first bottleneck structure is two-dimensional convolution, and the convolution layers of the rest bottleneck structures are one-dimensional convolution;
the output signal of the first bottleneck structure is input into the global average pooling layer and then sequentially input into the rest bottleneck structures;
the output signal of the last bottleneck structure is sequentially input to the attention module.
Further, the bottleneck structure includes 3 convolutional layers, 3 ReLU activation functions, and 1 downsampling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated output signals into the downsampling layer.
Further, the coarse positioning network comprises 3 convolutional layers, 3 ReLU activation functions, and 1 adaptive average pooling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated signals into the adaptive average pooling layer.
Further, the fine positioning network includes 5 convolutional layers, 5 ReLU activation functions, and 1 adaptive average pooling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated signals into the adaptive average pooling layer.
Further, the step of inputting the preprocessed chaotic ultrasonic signals into a preset neural network and outputting the position information of the key nodes of the component further comprises the steps of,
the neural network is trained using a sample dataset, wherein the sample dataset is divided into a simulated sample set and a real sample set.
The chaotic ultrasonic signals which are generated according to simulation software and are preprocessed are used as simulation samples in the simulation sample set, and the simulated position information of the chaotic ultrasonic signals is used as a label of the simulation samples;
taking the chaotic ultrasonic signal transmitted by the preprocessed preset ultrasonic generating device as an actual sampling sample in the actual sampling set, and taking the position information of the preset ultrasonic generating device as a label of the actual sampling sample;
dividing the simulation sample set and the actual sampling sample set into a simulation training set, a simulation test set, an actual sampling training set and an actual sampling test set according to a preset proportion;
Inputting the simulation training set and the corresponding label into the neural network for training, and inputting the simulation testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a first threshold;
and inputting the actual acquisition training set and the corresponding label into the trained neural network for training, inputting the actual acquisition testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a second threshold value, and outputting the neural network.
Further, the preprocessing the acquired chaotic ultrasonic signal includes,
supplementing 0 to the pre-generated chaotic ultrasonic signal to enable the length of the pre-generated chaotic ultrasonic signal to be the same as that of the acquired chaotic ultrasonic signal;
and splicing the chaotic ultrasonic signal subjected to 0 supplementation with the acquired chaotic ultrasonic signal with the same frequency band to obtain a preprocessed chaotic ultrasonic signal.
A specific description of each of the above embodiments is as follows.
The chaotic signal is a continuous, traversing, non-periodic signal. All samples in the traversal space will appear at least once. The aperiodicity of the chaotic signal can effectively avoid the matching error of the signal phase, and can also be used for filtering the non-main path signal component in the received signal, thereby avoiding the interference of multipath effect on positioning precision, improving the subsequent positioning performance, simultaneously, the chaotic signal has better phase difference information (time difference), avoiding the error generated by phase matching to cause time estimation to generate error so as to influence the positioning precision, and improving the accuracy of alarming.
There are various methods for generating the chaotic signal, and the analytic chaotic system is taken as an example for description. The analytic solution chaotic system is a chaotic system proposed by Corron N.J. et al in 2010, and a continuous time dynamics equation thereof can be represented by a differential equation (1):
wherein w represents the angular frequency of the mixed chaotic signal, x is a function of time t, is a continuous chaotic signal with amplitude changing along with time, is a one-dimensional signal, only needs to be transmitted according to the intensity of x (t) during transmission, gamma is a negative attenuation coefficient, and takes a value of 0 to ln2 in a left-right closed zone, and only the system in the range is chaotic. s=sgn (x (t)) is a switching function.
The switching condition of s is defined as:
That is, when the first derivative of x is 0, s (t) is assigned as sgn (x (t)), s is a symbol sequence, s is a value of ±1, s is changed when the signal derivative is 0, and the rest of time s (t) remains unchanged until the next switching condition is satisfied.
Deducing an analytical solution according to the formula (1) and the formula (2) under the condition that n is less than or equal to t is less than or equal to n+1,
s n is the sign value at t=n, x n The sampling value of the chaotic signal when t=n satisfies the following iterative relationship:
x n+1 ＝e γ x n -(e γ -1)x n
and analyzing the iteration type, wherein the optimal value of x (0) is-0.3776, so that a chaotic waveform can be generated and used for transmitting by the ultrasonic generating device.
In this embodiment, the key nodes of the component may be four corners where the surface to be butted is located, so as to accurately locate and reflect the posture condition and change of the component, and achieve the purpose of real-time warning of the abnormal state of the component.
Four or more ultrasonic generating devices are arranged on key nodes of the component, and the ultrasonic generating devices are combined with a chaotic system to generate chaotic ultrasonic signals, so that the ultrasonic generating devices respectively send the chaotic ultrasonic signals with different frequency bands.
Four or more ultrasonic acquisition devices are arranged at fixed positions in the component alignment and assembly site, and all the ultrasonic acquisition devices form an alarm ultrasonic acquisition device group.
Then, the acquired chaotic ultrasonic signals are processed, and 0 is supplemented to the pre-generated chaotic ultrasonic signals, so that the length of the pre-generated chaotic ultrasonic signals is the same as that of the acquired chaotic ultrasonic signals;
and splicing the chaotic ultrasonic signal subjected to 0 supplementation with the acquired chaotic ultrasonic signal with the same frequency band to obtain a preprocessed chaotic ultrasonic signal.
Specifically, the pretreatment operation can be divided into the following two parts:
(1) And (5) chaotic ultrasonic signal receiving.
Because the ultrasonic acquisition device cannot synchronously receive the signal data sent by the transmitting end, the receiving window of the receiving end is set to be larger than the data quantity sent by the transmitting end. For example, a 2 second signal is transmitted and received in a window of 3 seconds.
(2) And (5) splicing the chaotic ultrasonic signals.
Taking 4 paths as an example (4 ultrasonic acquisition devices), a computer (alarm data processing module) acquires 4 paths of signals (signals sent by the same ultrasonic generation device reach different ultrasonic acquisition devices through different paths and are acquired) of the same frequency band received by the alarm ultrasonic acquisition device group, and chaotic ultrasonic signals of the same frequency band come from the same ultrasonic generation module. Because the duration of the receiving window is longer than that of the transmitted chaotic ultrasonic signal, the data length of the received chaotic ultrasonic signal is longer than that of the pre-generated chaotic ultrasonic signal for transmission, and therefore the rear end of the pre-generated chaotic ultrasonic signal needs to be spliced with the four paths of same-frequency signals after being subjected to 0 compensation. For example, the data shape of the received chaotic ultrasonic signal and the pre-generated chaotic ultrasonic signal after 0 supplement is [1, 20000], and the data shape after splicing is [5, 20000], wherein 5 represents the signal received by 4 ultrasonic acquisition devices and the signal after 0 supplement of 1 pre-generated chaotic ultrasonic signal, and 20000 represents the information received by each ultrasonic acquisition device.
Referring to fig. 2, the structure of the neural network of the present application includes a backbone network, a coarse positioning network, and a fine positioning network. After the preprocessed signals are input into the backbone network, the output signals of the backbone network are input into the coarse positioning network and the fine positioning network, and the output signals of the coarse positioning network are input into the fine positioning network.
Referring to fig. 3, the backbone network structure consists of five bottleneck structures, one global averaging pooling layer, and two attention modules. The attention module includes a channel attention module and a spatial attention module. The two attention modules can enable the neural network to pay more attention to important information, and reduce attention to local invalid information, so that the purposes of simplifying a model and accelerating calculation are achieved.
Wherein, the convolution layer of the first bottleneck structure is two-dimensional convolution, and the convolution layers of other bottleneck structures are one-dimensional convolution. The global averaging pooling layer maps the output of the first bottleneck structure to [1,10000], which is then sequentially input into the remaining bottleneck structures.
Referring to fig. 3-4, the number of convolution kernels C, the convolution kernel size K, and the value P complementary to the convolution process of the bottleneck structure are set. In the first bottleneck structure (i.e., bottleneck structure-1), C32K5 represents a convolution kernel of 32 5x5, and in the remaining bottleneck structures, K is fixedly set to 1x9 and p is set to 4, so as to prevent data from missing boundary data and supplementing the data due to convolution characteristics.
The final bottleneck structure output result is multiplied by the channel attention module and then is multiplied by the space attention module, and then is output, and the final output size of the backbone network is [512,625].
Referring to fig. 4, the bottleneck structure includes 3 convolutional layers, 3 ReLU activation functions, and 1 downsampling layer.
Referring to fig. 5, the coarse positioning network structure includes 3 convolutional layers, 3 ReLU activation functions, and 1 adaptive average pooling layer. The coarse positioning network receives the output of the backbone network and outputs a vector of shape [1,3 ]. Three elements in the vector are respectively coarse positioning space coordinates of key nodes, and are expressed by (x, y, z) coordinates, and the precision is in the centimeter level.
Referring to fig. 6, the fine positioning network structure contains 5 convolutional layers, 5 ReLU activation functions, and 1 adaptive average pooling layer. The fine positioning network receives a matrix formed by splicing the output of the main network and the coarse positioning network. The shape of the matrix is [515,625], the front 512 acts as the output matrix of the backbone network, and the back 3 acts are repeated by 3 coarsely positioned space coordinates. The output of the fine positioning network is a matrix of 1 [3, 11 ]. Where the location of the maximum element represents the output of the fine positioning. The 11 elements of the output vector respectively correspond to integers of 5 to 5, and represent the correction of the positioning output result of the coarse positioning neural network under millimeter precision.
The characteristic information is extracted through the main network of the neural network, the coarse positioning network of the neural network outputs coordinates, and the coarse positioning network can output the position information of the key nodes through effective training due to the fact that the precision is in the centimeter level.
The fine positioning network of the neural network receives the position information output by the coarse positioning network and the characteristic information extracted by the main network, and the effective correction of the positioning error of the coarse positioning network is realized by classifying the offset under the centimeter precision.
Specifically, assuming that the output of the coarse positioning network is CP and the output of the fine positioning network is FP, the final output result P of the neural network is:
P＝[P X ，P Y ，P Z ]
wherein P is X 、P Y 、P Z The coordinates of P on different coordinate axes are respectively obtained by the following formulas:
P X ＝CP X +0.1×FP X
P Y ＝CP Y +0.1×FP Y
P Z ＝CP Z +0.1×FP Z
wherein, CP and FP are three-dimensional coordinate forms:
CP＝[CP X ，CP Y ，CP Z ]
FP＝[FP X ，FP Y ，FP Z ]
wherein, FP X 、FP Y And FP Z Respectively [3, 11 ] of fine positioning network output]The position numbers of the maxima in the first, second and third rows of the matrix.
Further, when an alarm task with low accuracy requirement is executed and the accuracy requirement is only in a centimeter level range, the fine positioning network of the neural network can not work and output a result directly through the coarse positioning network of the neural network.
Further, the loss function of the neural network can evaluate the quality of the predicted value of the neural network, namely the degree of the difference between the predicted value and the true value. In each training process, inputting a predicted value of the neural network and a label corresponding to the data into the loss function, calculating an error value of the predicted value of the neural network and the label, and optimizing the neural network by utilizing the error value.
In training, the coarse positioning network adopts a mean square error as a loss function; the fine positioning network uses cross entropy as a loss function. Wherein, the formula of the mean square error is as follows:
wherein k is the sample data number; label coarse Training labels for coarse positioning networks; output put coarse Is the output of the coarse positioning network.
The formula of the cross entropy is as follows:
wherein the method comprises the steps ofK is the sample data number, label fine Training tags for a fine positioning network; output put fine For fine positioning of the network output.
Further, training of the neural network relies on the composition of the sample set, which is composed as follows:
the sample set required for training of the neural network includes a simulation sample set. The simulation sample set is generated by simulation software according to a digital model of an actual component assembly site and an ultrasonic generating device model randomly placed in the digital model. The label corresponding to each simulation sample is used for generating the position information of the ultrasonic generating device model of the simulation sample in the digital model of the component assembly site, and the position information is described by (x, y, z) coordinates. The simulation sample set comprises a data set acquired under the condition that the direct projection is not blocked and a data set acquired under the condition that a small part of direct projection paths are blocked by the model. The simulation sample set is divided into a simulation training set and a simulation test set according to the ratio of 8:2, and the corresponding labels are divided into a positioning (centimeter-level precision) form and an offset form, and are respectively used as the labels of the coarse positioning network and the fine positioning network. Wherein, the simulation software can use matlab and cad.
The sample set required for training the neural network also comprises an actual sample set. The actual sampling sample set is acquired by an alarm ultrasonic acquisition device group of an alarm system and a data preprocessing sub-module in an alarm data processing module. The label corresponding to each actual sampling sample is the actual position of the ultrasonic generating device for generating the actual sampling sample in the component assembly site, and is described by (x ', y ', z ') coordinates. The real sampling set comprises a data set collected under the condition that the signal is directly projected and is not shielded under the real sampling condition in the field and a data set collected under the condition that a small part of direct paths are shielded. The actual acquisition sample set is divided into an actual acquisition training set and an actual acquisition test set according to the ratio of 8:2, and the corresponding labels are divided into a form of positioning (centimeter-level precision) and offset, and are respectively used as a label of a coarse positioning network and a label of a fine positioning network.
The data format of the samples and tags in the simulation sample set is completely consistent with the data format in the real sample set.
Finally, training the neural network. The training of the neural network comprises two stages, namely a simulation training stage and an actual acquisition training stage.
The two-stage training process is completed by using different training set samples, namely, after the neural network is trained by using the simulation sample set, the network is trained by using the real sampling set continuously. The training process comprises the following steps:
A. Inputting the training set samples into a neural network, the neural network producing two predicted outputs: coarse positioning results and fine positioning results;
B. the two output results are firstly respectively calculated into a loss function L with divided positioning information (centimeter-level precision) and offset in the corresponding training sample label coares And L fine ；
C. Using coarse positioning network loss function L coarse And a fine positioning network loss function L fine Updating the weight of the corresponding neural network by a gradient descent method;
D. the above steps A, B and C are repeated continuously,
and in the simulation sample set training stage, after training on the simulation training set for a period of time, stopping the training when the test accuracy rate on the simulation test set reaches a first threshold value (100%), and entering the real sample set training stage. Similarly, after the training of the actual acquisition training set for a period of time, stopping the training when the test accuracy rate on the actual acquisition test set reaches a second threshold value (98%), and completing the training of the neural network.
The trained deep neural network is adopted to receive the chaotic ultrasonic signals subjected to data preprocessing, namely the position information of the key nodes of the component can be output, different speed thresholds and boundary thresholds can be set according to the tasks of butt joint installation of different components, whether the position boundary and the movement speed of the key nodes of the component or the assembly component exceed the preset boundary and threshold is judged through the position information of the key nodes output by the neural network, and the alarm is given if the position boundary and the movement speed of the key nodes of the component or the assembly component exceed the preset boundary and threshold.
In summary, in the alarming method applied to the in-field component or part alignment assembly process of large and ultra-large components, by acquiring chaotic ultrasonic signals of different frequency bands sent out by different key nodes of the components, processing the acquired chaotic ultrasonic signals by using a neural network and learning and outputting spatial coordinate information of ultrasonic generating devices at different positions on the components, so as to obtain the position information of the key nodes, judging whether the position boundary and the movement speed of the key nodes exceed preset boundaries and thresholds by using the position information of the outputted key nodes, and alarming if the position boundary and the movement speed of the key nodes exceed the preset boundaries and thresholds. The chaotic ultrasonic signals in different frequency bands are not interfered with each other, so that crosstalk between the chaotic ultrasonic signals is reduced, the distance measurement information contained in the preprocessing spliced signals is ensured to come from the same ultrasonic generating device, and the neural network is beneficial to accurately outputting the position information of the key node; the situation that errors are generated due to phase matching caused by the periodicity of signals, and positioning accuracy is affected is improved. The possibility of false alarm and non-alarm of an alarm system is reduced; compared with the traditional positioning algorithm, the neural network output key node position information has better anti-interference capability, and can output more accurate position information under the condition that the input signal contains a small amount of interference, so that the probability of false alarm and non-alarm occurrence of an alarm system is effectively reduced; the situation that the direct path of signals is blocked by an actuating mechanism or an assembly part in the field to cause false alarm in the traditional alarm method process is improved.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic of each process, and should not limit the implementation process of the embodiment of the present application in any way.
Referring to fig. 7, the embodiment of the application further provides an alarm system, which includes an ultrasonic generating module, an alarm ultrasonic acquisition device group, an alarm data processing module, an alarm judging module and an alarm module;
the ultrasonic generating module comprises four or more ultrasonic generating devices, the specific number of the ultrasonic generating devices is determined by the structure of a component or a mounting part, in the embodiment, the number of the ultrasonic generating devices can be four, the ultrasonic generating devices are arranged on key nodes of the component, ultrasonic chaotic signals which are not mutually interfered are respectively emitted on different frequency bands, and the signal frequencies of the ultrasonic generating devices at the same position are the same;
the alarming ultrasonic acquisition device group comprises four or more ultrasonic acquisition devices, the ultrasonic acquisition devices are arranged at fixed positions in the field, which are close to the working field, and each ultrasonic acquisition device is provided with a plurality of signal acquisition channels for receiving chaotic ultrasonic signals of corresponding frequency bands;
The number of ultrasonic acquisition devices depends on the probability that the mechanical arm or other actuator of the industrial robot blocks the direct signal path in different situations. If the number of the ultrasonic acquisition devices is less than four, the received information is not enough for the neural network learning, namely the neural network is difficult to learn effective information. If the mechanical arm or other actuating mechanism of the industrial robot is dense in the field space where the components or the assembly parts are aligned and installed, the probability of blocking one path of signal from directly radiating paths is high, the number of ultrasonic acquisition devices needs to be increased to provide more path ranging information for the neural network to support positioning. The number of the ultrasonic acquisition devices is too large, the processing capacity of data can be increased, and the instantaneity of the output position information of the neural network is reduced, so that the number of the ultrasonic acquisition devices is not too large. In this embodiment, four ultrasound acquisition devices are selected. When the probability that a mechanical arm or other executing mechanisms of the industrial robot in the field block the direct signal path is small, the four ultrasonic acquisition devices can finish accurate positioning, and the instantaneity of outputting positioning information is also good;
the alarm data processing module is in communication connection with the alarm ultrasonic acquisition device group and is used for receiving chaotic ultrasonic signals with different frequencies, and splicing the chaotic signals received with the same frequency band to generate spliced signals corresponding to the number of ultrasonic generation devices;
The alarm judging module is connected with the output end of the alarm data processing module and is used for outputting the position information of the key node of the component through a preset neural network and judging whether an alarm is given or not based on the position information and a set threshold value;
specifically, the neural network performs reasoning calculation based on the spliced signals to obtain position information of an ultrasonic generating device which corresponds to the spliced signals and is positioned at a key node of a component or an installation part; in this embodiment, the position information of the four ultrasonic generating devices is included; neural networks are mathematical models that mimic animal neural network systems and have learning capabilities. In training, the neural network calculates an output result according to the input sample, and adjusts internal parameters according to the output result and the error of the sample label. Thus, the neural network can learn how to go from data to the desired result in the training process without manually designing the calculation flow thereof;
specifically, whether to alarm is judged based on the position information and the set threshold value, different speed threshold values and preset boundaries can be set according to the task of docking or installing different components, whether the position boundary and the movement speed of the key node of the component exceed the preset boundary and speed threshold values or not is judged according to the position information of the key node output by the neural network, and an alarm is given if the position boundary and the movement speed of the key node of the component exceed the preset boundary and speed threshold values, namely an alarm instruction is output backwards;
The alarm module is connected to the output end of the alarm judging module and used for comprehensively alarming according to the alarm instruction output by the alarm judging module; the alarm module can be alarm facilities such as an alarm horn, an alarm lamp and the like in the field.
Referring to fig. 8, the generation of the chaotic ultrasonic signal of the ultrasonic generating device, the data processing of the chaotic ultrasonic signal acquired by the alarm ultrasonic acquisition device group and the judgment of the alarm state are all completed by the same computer system.
In summary, an alarm system has four or more ultrasound generating devices disposed at key nodes of a component. These devices transmit ultrasonic chaotic signals of different frequency bands.
Four or more ultrasonic acquisition devices are arranged at fixed positions of a site, so that an alarm system comprises redundant ultrasonic acquisition devices, and the redundancy can provide more position reasoning basis for the deep neural network, so that the deep neural network can output more accurate positioning; and because the spliced signal contains redundant multipath ranging information, when one path of signal is blocked, the information contained in other paths in the spliced signal is enough to realize accurate positioning.
The neural network is utilized to process the received and preprocessed chaotic ultrasonic signals, the position information of the key node where the corresponding ultrasonic generating device is located is output, whether the position boundary and the movement speed of the component or the assembly part exceed the preset boundary and threshold value is judged according to the output position information, and when the boundary and the speed exceed the set threshold value, an alarm is given. The neural network has good anti-interference capability, can learn how to extract useful information from noise or multipath interference conditions to position in the training process, can carry out self-correction through data driving in the training process, does not need a large number of extra experiments to compensate and correct an algorithm model, has higher positioning accuracy than the traditional ultrasonic positioning algorithm under the condition of noise and other interference, and further reduces the possibility of false alarm and alarm failure.
Because the data set contains a plurality of data and label pairs acquired when the direct wave is blocked, and redundant ranging information is provided, the data set can learn the information extraction capability through the neural network, and can realize accurate positioning by using other path information when one path of signals are blocked after multiple rounds of training, so that the possibility of false alarm occurrence is reduced.
And alarming facilities such as alarming horns, alarming lamps and the like in the field carry out alarming after receiving the alarming instruction.
Therefore, the alarm system is applied to the near field alignment of large and ultra-large components or the installation of accessory components in the component assembly process, and the situation that the direct path of signals is blocked by an in-site actuating mechanism or an assembly component to cause false alarm in the traditional alarm process is avoided; and the positioning accuracy is improved, and the system false alarm or alarm condition caused by inaccurate positioning of key nodes when the traditional positioning algorithm is interfered by noise and other conditions is effectively reduced.
Specific limitations regarding an alarm system may be found in the foregoing description of an alarm method, and are not described in detail herein.
Each of the modules in an alarm system described above may be implemented in whole or in part by software, hardware, and combinations thereof. The above modules may be embedded in hardware or may be independent of a processor in the computer device, or may be stored in software in a memory in the computer device, so that the processor may call and execute operations corresponding to the above modules.
In one embodiment, a computer device is provided, which may be a server. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements any of the alert methods described above.
In one embodiment, a computer readable storage medium is provided, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the steps of when executing the computer program:
s1, acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of a component;
s2, preprocessing the acquired chaotic ultrasonic signals;
S3, inputting the preprocessed chaotic ultrasonic signals into a preset neural network, and outputting position information on key nodes of a component;
and S4, judging whether to alarm or not based on the position information and the set threshold value.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in the various embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous Link DRAM (SLDRAM), memory bus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the system is divided into different functional units or modules to perform all or part of the above-described functions.
Claims (7)
1. An alarm method, characterized by being applied to the alignment and assembly process of large and ultra-large components or accessory components, comprising the following steps,
acquiring chaotic ultrasonic signals of different frequency bands transmitted on key nodes of a component, wherein the chaotic system generates chaotic waveforms for the ultrasonic generation module to transmit, and the chaotic ultrasonic signals are signals transmitted by the ultrasonic generation module and the signal amplitude of which is determined by the waveform amplitude of the chaotic system, so that the ultrasonic generation modules positioned on different key nodes of the component respectively transmit the chaotic ultrasonic signals of different frequency bands;
preprocessing the acquired chaotic ultrasonic signals;
inputting the preprocessed chaotic ultrasonic signals into a preset neural network, and outputting the position information of key nodes of a component;
Judging whether to alarm or not based on the position information and the set threshold value;
the neural network comprises a backbone network, a coarse positioning network and a fine positioning network;
the output signals of the backbone network are input into the coarse positioning network and the fine positioning network;
the output signal of the coarse positioning network is input into the fine positioning network;
the backbone network consists of five bottleneck structures, a global average pooling layer and two attention modules;
wherein, the convolution layer of the first bottleneck structure is two-dimensional convolution, and the convolution layers of the rest bottleneck structures are one-dimensional convolution;
the output signal of the first bottleneck structure is input into the global average pooling layer and then sequentially input into the rest bottleneck structures;
the output signal of the last bottleneck structure is sequentially input into the attention module;
the bottleneck structure comprises 3 convolution layers, 3 ReLU activation functions and 1 downsampling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
accumulating the output signals of the ReLU activation function and then inputting the accumulated signals into the downsampling layer;
the coarse positioning network comprises 3 convolution layers, 3 ReLU activation functions and 1 adaptive average pooling layer;
The convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
accumulating the output signals of the ReLU activation function and then inputting the accumulated signals into the self-adaptive average pooling layer;
the fine positioning network comprises 5 convolution layers, 5 ReLU activation functions and 1 adaptive average pooling layer;
the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output;
and accumulating the output signals of the ReLU activation function and inputting the accumulated signals into the adaptive average pooling layer.
2. The warning method of claim 1 wherein the step of determining whether to alert based on the location information and the set threshold comprises,
acquiring position information of key nodes on a component;
and judging whether the position boundary of the key node exceeds a preset boundary range and whether the movement speed exceeds a threshold value according to the position information of the key node, and alarming if the position boundary exceeds the threshold value.
3. The warning method of claim 1, wherein the step of inputting the preprocessed chaotic ultrasonic signal into a preset neural network, outputting position information of key nodes of a component further comprises,
Training the neural network by using a sample data set, wherein the sample data set is divided into a simulation sample set and a real sample set;
the chaotic ultrasonic signals which are generated according to simulation software and are preprocessed are used as simulation samples in the simulation sample set, and the simulated position information of the chaotic ultrasonic signals is used as a label of the simulation samples;
taking the chaotic ultrasonic signal transmitted by the preprocessed preset ultrasonic generating device as an actual sampling sample in the actual sampling set, and taking the position information of the preset ultrasonic generating device as a label of the actual sampling sample;
dividing the simulation sample set and the actual sampling sample set into a simulation training set, a simulation test set, an actual sampling training set and an actual sampling test set according to a preset proportion;
inputting the simulation training set and the corresponding label into the neural network for training, and inputting the simulation testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a first threshold;
and inputting the actual acquisition training set and the corresponding label into the trained neural network for training, inputting the actual acquisition testing set into the trained neural network for testing after each training until the testing accuracy of the neural network reaches a second threshold value, and outputting the neural network.
4. The method of claim 1, wherein the step of preprocessing the acquired chaotic ultrasonic signal comprises,
supplementing 0 to a chaotic ultrasonic signal pre-generated by a chaotic system, so that the length of the chaotic ultrasonic signal is the same as that of the acquired chaotic ultrasonic signal;
and splicing the chaotic ultrasonic signal subjected to 0 supplementation with the acquired chaotic ultrasonic signal with the same frequency band to obtain a preprocessed chaotic ultrasonic signal.
5. An alarm system is characterized by comprising an ultrasonic generation module, an alarm ultrasonic acquisition device group, an alarm data processing module, an alarm judging module and an alarm module;
the ultrasonic generation modules are respectively arranged on key nodes of the component and used for transmitting chaotic ultrasonic signals with different frequency bands, wherein the chaotic system generates chaotic waveforms for the ultrasonic generation modules to transmit, and the chaotic ultrasonic signals are signals transmitted by the ultrasonic generation modules and the signal amplitudes of which are determined by the waveform amplitudes of the chaotic system;
the alarming ultrasonic acquisition device group is used for acquiring chaotic ultrasonic signals of different frequency bands transmitted from key nodes of the component;
the alarm data processing module is in communication connection with the alarm ultrasonic acquisition device group and is used for preprocessing the acquired chaotic ultrasonic signals;
The alarm judging module is used for outputting the position information of the key nodes of the component through a preset neural network and judging whether an alarm is given or not based on the position information and a set threshold value; the neural network comprises a backbone network, a coarse positioning network and a fine positioning network; the output signals of the backbone network are input into the coarse positioning network and the fine positioning network; the output signal of the coarse positioning network is input into the fine positioning network; the backbone network consists of five bottleneck structures, a global average pooling layer and two attention modules; wherein, the convolution layer of the first bottleneck structure is two-dimensional convolution, and the convolution layers of the rest bottleneck structures are one-dimensional convolution; the output signal of the first bottleneck structure is input into the global average pooling layer and then sequentially input into the rest bottleneck structures; the output signal of the last bottleneck structure is sequentially input into the attention module; the bottleneck structure comprises 3 convolution layers, 3 ReLU activation functions and 1 downsampling layer; the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output; accumulating the output signals of the ReLU activation function and then inputting the accumulated signals into the downsampling layer; the coarse positioning network comprises 3 convolution layers, 3 ReLU activation functions and 1 adaptive average pooling layer; the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output; accumulating the output signals of the ReLU activation function and then inputting the accumulated signals into the self-adaptive average pooling layer; the fine positioning network comprises 5 convolution layers, 5 ReLU activation functions and 1 adaptive average pooling layer; the convolution layer and the ReLU activation function are arranged in a matched mode, and an output signal of the convolution layer is input into the ReLU activation function and then output; accumulating the output signals of the ReLU activation function and then inputting the accumulated signals into the self-adaptive average pooling layer;
And the alarm module is used for automatically alarming based on the judging result of the alarm judging module.
6. A computer device comprising a memory, a processor and a computer program stored on the memory, the processor executing the computer program to perform the steps of the method of any one of claims 1 to 4.
7. A computer-readable storage medium, characterized in that it stores a computer program which, when executed by a processor, implements the steps of the method of any one of claims 1 to 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211269854.0A CN115713844B (en) | 2022-10-18 | 2022-10-18 | Alarm method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211269854.0A CN115713844B (en) | 2022-10-18 | 2022-10-18 | Alarm method and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115713844A CN115713844A (en) | 2023-02-24 |
| CN115713844B true CN115713844B (en) | 2023-06-23 |
Family
ID=85229932
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211269854.0A Active CN115713844B (en) | 2022-10-18 | 2022-10-18 | Alarm method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115713844B (en) |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20000067177A (en) * | 1999-04-24 | 2000-11-15 | 유창모 | Sonic insect repeller using chaotic phenomenon |
| JP3534667B2 (en) * | 1999-11-17 | 2004-06-07 | 義文 原田 | Ultrasonic measuring device |
| CN103336054B (en) * | 2013-06-03 | 2016-03-02 | 北京工业大学 | Based on the butt-weld lossless detection method of ultrasonic Lamb wave |
| CN104155364A (en) * | 2014-07-31 | 2014-11-19 | 暨南大学 | Method and device for identifying defect position of pipeline based on phase locus partitioned by round window |
| CN105116371B (en) * | 2015-06-30 | 2017-08-29 | 华南理工大学 | A kind of object localization method and device based on Continuous Transmission Frequency Modulation signal |
| CN105954358B (en) * | 2016-05-04 | 2018-09-18 | 中国计量大学 | A kind of small defect location detection method of supersonic guide-wave that TR is combined with Duffing systems |
| CN106959157B (en) * | 2017-04-17 | 2019-05-28 | 燕山大学 | A kind of ultrasonic detection method based on Duffing and SH-COS chaos system |
| CN110274956A (en) * | 2019-07-10 | 2019-09-24 | 广东海洋大学 | A kind of pipe ultrasonic guided wave detecting method based on Kolmogorov entropy |
| CN114184682B (en) * | 2021-12-10 | 2024-04-30 | 东莞理工学院 | Double-chaotic system detection method for weak ultrasonic guided wave signals |
| CN115436881B (en) * | 2022-10-18 | 2023-07-07 | 兰州大学 | Positioning method, positioning system, computer equipment and readable storage medium |
-
2022
- 2022-10-18 CN CN202211269854.0A patent/CN115713844B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN115713844A (en) | 2023-02-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP2021506149A (en) | Motion detection based on machine learning of radio signal characteristics | |
| US10275707B2 (en) | Systems and methods for training multipath filtering systems | |
| US5537511A (en) | Neural network based data fusion system for source localization | |
| CN111008364A (en) | Method and system for cooperative passive positioning of double observers | |
| US5488589A (en) | Neural network based three dimensional ocean modeler | |
| Efimov et al. | Artificial neural network based angle-of-arrival estimator | |
| CN115713844B (en) | Alarm method and system | |
| CN115436881B (en) | Positioning method, positioning system, computer equipment and readable storage medium | |
| Baum | Simultaneous tracking and shape estimation of extended objects | |
| Liu et al. | General spiking neural network framework for the learning trajectory from a noisy mmwave radar | |
| Le Gentil et al. | Accurate Gaussian-Process-based Distance Fields with Applications to Echolocation and Mapping | |
| Chen et al. | Research on AIS and radar information fusion method based on distributed Kalman | |
| CN109901189B (en) | Three-dimensional point cloud tracking device and method using recurrent neural network | |
| US20220103272A1 (en) | Device and method for assessing a state of a radio channel | |
| Sporyshev et al. | Reinforcement learning approach for cooperative AUVs in underwater surveillance operations | |
| US6009185A (en) | Neural network based contact state estimator | |
| Smith et al. | Stochastic modeling and control for tracking the periodic movement of marine animals via AUVs | |
| Rogers | The Fornasini-Marchesini model—Its role in the analysis and control of physical systems | |
| Godsill | Particle filters for continuous-time jump models in tracking applications | |
| Bucurica et al. | Virtual environment using cellular automata for experimental studies of sound propagation and echo classification with artificial neural networks | |
| Pan et al. | Deconvolved conventional beamforming and adaptive cubature Kalman filter based distant speech perception system | |
| JP7328885B2 (en) | MOTION SENSOR SELECTION DEVICE, TARGET DETECTION SYSTEM AND MOTION SENSOR SELECTION METHOD | |
| CN117911852B (en) | Underwater target distance prediction method based on self-adaption in part of unsupervised field | |
| Goldfrank et al. | Risk Reduction in Target Motion Analysis Using Approximate Dynamic Programming | |
| Aihara et al. | Nonlinear dynamics in free flight of an echolocating bat |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP02 | Change in the address of a patent holder | ||
| CP02 | Change in the address of a patent holder |
Address after: 730070 No. 118 Anning West Road, Anning District, Gansu, Lanzhou Patentee after: Lanzhou Jiaotong University Address before: No. 118, Anning West Road, Anning District, Guangzhou City, Guangdong Province Patentee before: Lanzhou Jiaotong University |