CN113537160A - Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium - Google Patents

Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium Download PDF

Info

Publication number
CN113537160A
CN113537160A CN202111065545.7A CN202111065545A CN113537160A CN 113537160 A CN113537160 A CN 113537160A CN 202111065545 A CN202111065545 A CN 202111065545A CN 113537160 A CN113537160 A CN 113537160A
Authority
CN
China
Prior art keywords
data
ball mill
vibration
audio
signal data
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.)
Granted
Application number
CN202111065545.7A
Other languages
Chinese (zh)
Other versions
CN113537160B (en
Inventor
陈桂刚
宋晨
袁晓艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tianjin Zhongxin Zhiguan Information Technology Co ltd
Original Assignee
Tianjin Zhongxin Zhiguan Information Technology Co ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tianjin Zhongxin Zhiguan Information Technology Co ltd filed Critical Tianjin Zhongxin Zhiguan Information Technology Co ltd
Priority to CN202111065545.7A priority Critical patent/CN113537160B/en
Publication of CN113537160A publication Critical patent/CN113537160A/en
Application granted granted Critical
Publication of CN113537160B publication Critical patent/CN113537160B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B02—CRUSHING, PULVERISING, OR DISINTEGRATING; PREPARATORY TREATMENT OF GRAIN FOR MILLING
    • B02C—CRUSHING, PULVERISING, OR DISINTEGRATING IN GENERAL; MILLING GRAIN
    • B02C17/00—Disintegrating by tumbling mills, i.e. mills having a container charged with the material to be disintegrated with or without special disintegrating members such as pebbles or balls
    • B02C17/18—Details
    • B02C17/1805—Monitoring devices for tumbling mills
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01H—MEASUREMENT OF MECHANICAL VIBRATIONS OR ULTRASONIC, SONIC OR INFRASONIC WAVES
    • G01H17/00—Measuring mechanical vibrations or ultrasonic, sonic or infrasonic waves, not provided for in the preceding groups
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
    • G01M99/00—Subject matter not provided for in other groups of this subclass
    • G01M99/007—Subject matter not provided for in other groups of this subclass by applying a load, e.g. for resistance or wear testing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/03—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters
    • G10L25/24—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the type of extracted parameters the extracted parameters being the cepstrum
    • G—PHYSICS
    • G10—MUSICAL INSTRUMENTS; ACOUSTICS
    • G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
    • G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
    • G10L25/27—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique
    • G10L25/30—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 characterised by the analysis technique using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Data Mining & Analysis (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Signal Processing (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Audiology, Speech & Language Pathology (AREA)
  • Human Computer Interaction (AREA)
  • Acoustics & Sound (AREA)
  • Multimedia (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Food Science & Technology (AREA)
  • Evolutionary Biology (AREA)
  • Measurement Of Mechanical Vibrations Or Ultrasonic Waves (AREA)

Abstract

The invention provides a method and a device for measuring load of a ball mill, electronic equipment and a medium, which relate to the technical field of ore grinding, and the method comprises the following steps: acquiring vibration signal data and audio signal data of the ball mill; performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data; the method comprises the steps of obtaining a pre-trained multi-modal neural network model, inputting vibration IMF component characteristic data and audio mfcc characteristic data into the multi-modal neural network model (comprising a characteristic extraction network (a vibration signal CNN sub-network and an audio signal CNN sub-network) and a two-layer fully-connected BP neural network connected with the characteristic extraction network), and obtaining a ball mill load measurement result. The invention can accurately and effectively measure the filling amount of the ball mill, provides possibility for real-time control of the filling amount of the ball mill, can improve the working efficiency of the ball mill, and reduces the power consumption and the raw material consumption.

Description

Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium
Technical Field
The invention relates to the technical field of ore grinding, in particular to a method and a device for measuring load of a ball mill, electronic equipment and a medium.
Background
The variation of mill load (i.e., mill fill) control is complicated and difficult due to the long hysteresis characteristics of the mill and the physical properties of the mill feed material (such as variations in particle size, moisture, and grindability, as well as variations in mill backing and grinding media).
At present, when the load of the mill is measured, the measurement of the load of the mill is difficult due to the noise influence of the working environment and the change of the working condition. The traditional method is based on a mode of listening to sound manually, but the measurement accuracy cannot meet the requirement, so the quality of the product is influenced, the power consumption is increased, and raw materials are wasted. Although there are many methods to optimize mill fill level measurements, such as improving the problem of mounting sensors, and using algorithmic optimization, as well as multi-sensor fusion methods, none have fully addressed the above-mentioned effects of noise.
Disclosure of Invention
The invention aims to provide a ball mill load measuring method, a ball mill load measuring device, electronic equipment and a medium, which can accurately and effectively measure the filling amount of a ball mill, provide possibility for real-time control of the filling amount of the ball mill, improve the working efficiency of the ball mill and reduce the power consumption and the raw material consumption.
In a first aspect, the present invention provides a method for measuring a load of a ball mill, the method comprising: acquiring vibration signal data and audio signal data of the ball mill; performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data; acquiring a pre-trained multi-modal neural network model, and inputting vibration IMF component characteristic data and audio mfcc characteristic data into the multi-modal neural network model to obtain a ball mill load measurement result; the pre-trained multi-mode neural network model comprises a feature extraction network and a two-layer fully-connected BP neural network connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
In an optional embodiment, the pre-trained multi-modal neural network model is trained in the cloud, and the training step of the pre-trained multi-modal neural network model includes: acquiring label data and characteristic data; the characteristic data comprises vibration IMF component characteristic data and audio mfcc characteristic data; performing data enhancement processing on the characteristic data, and performing data truncation processing on the vibration IMF component characteristic data and the audio mfcc characteristic data after the data enhancement processing to obtain a plurality of truncated data; adding each truncated data serving as a sample into training sample data, and adding white noise data into the training sample data to perform data expansion processing to obtain target training sample data; and training the multi-modal neural network model by adopting target training sample data.
In an alternative embodiment, the sub-network of the vibration signal CNN includes a batch normalization layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, and a batch normalization layer connected in sequence; the audio signal CNN sub-network comprises a batch standardization layer, an 11 × 11 convolution layer, a pooling layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, a full connection layer and a batch standardization layer which are connected in sequence.
In an alternative embodiment, the step of acquiring vibration signal data and audio signal data of the ball mill comprises: acquiring vibration signal data and audio signal data of the ball mill acquired by an acquisition device based on a preset acquisition frequency to obtain initial acquisition data; sampling and band-pass filtering the initial acquisition data; wherein the first filtering range for the vibration signal data and the second filtering range for the audio signal data are different.
In an optional implementation manner, the step of performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data includes: performing empirical mode decomposition of a preset number on the vibration signal data to obtain vibration IMF component characteristic data; and performing mfcc feature extraction on the audio signal data to obtain audio mfcc feature data.
In an optional embodiment, the step of performing a preset number of empirical mode decompositions on the vibration signal data to obtain vibration IMF component characteristic data includes: step S1, determining all local maximum value points and local minimum value points of the vibration signal data, and fitting by a cubic spline interpolation function to form an upper envelope line and a lower envelope line of the original data; step S2, calculating the mean of the upper envelope and the lower envelope
Figure F_210909172728492_492083001
Determining a new data sequence
Figure F_210909172728729_729022002
(ii) a New data sequence
Figure F_210909172728997_997444003
To remove sequences of low frequencies; step S3, judging new data sequence
Figure F_210909172729248_248413004
Whether the condition that IMF is established is satisfied, and if so, determining
Figure F_210909172729468_468653005
Repeating the steps S1 and S2 if the first IMF component of the vibration signal data is not satisfied, obtaining
Figure F_210909172729754_754697006
(ii) a Wherein the content of the first and second substances,
Figure M_210909172739191_191276001
a new data sequence confirmed after the first iteration;
Figure M_210909172739222_222515002
the mean value of the upper envelope line and the lower envelope line obtained after the first repeated operation; step S4, judgment
Figure F_210909172730007_007718007
If the IMF is satisfied, repeating the steps S1 to S3 k times until the IMF is satisfied
Figure F_210909172730292_292902008
，
Figure F_210909172730625_625331009
Finally, the IMF condition is satisfied, and the IMF condition is the first IMF component and is recorded as:
Figure F_210909172730864_864164010
(ii) a Wherein the content of the first and second substances,
Figure M_210909172739253_253818003
a new data sequence confirmed after the kth repeat operation;
Figure M_210909172739302_302236004
a new data sequence confirmed after the k-1 repeated operation;
Figure M_210909172739333_333861005
the mean value of the upper envelope line and the lower envelope line obtained after the kth repeated operation;
Figure M_210909172739365_365091006
is the first IMF component; step S5, subtracting the vibration signal data
Figure F_210909172731148_148370011
Obtaining a first residual component
Figure F_210909172731403_403727012
Will be
Figure F_210909172731688_688854013
As the signal to be decomposed, the above steps S1 to S4 are performed to obtain the second IMF component
Figure M_210909172739443_443220007
And determining a second residual component
Figure M_210909172739492_492986008
Repeating the above steps S1 to S4 until the nth IMF component is obtained
Figure M_210909172739524_524252009
And an nth residual component to obtain n IMF components after decomposing the vibration signal data
Figure M_210909172739571_571144010
And n residual components
Figure M_210909172739602_602543011
。
In an alternative embodiment, the step of performing mfcc feature extraction on the audio signal data to obtain audio mfcc feature data includes: performing framing processing on the audio signal data to obtain a plurality of frame data; frame data between two adjacent frames contain repeated data; performing digital signal processing on frame data of each frame; the digital signal processing comprises adding a Hamming window, Fourier transformation, calculating a power spectrum and triangular band-pass filtering; and calculating the logarithmic energy output by the filter according to the frame data after the digital signal processing, and performing discrete cosine transform to obtain audio mfcc characteristic data.
In a second aspect, an embodiment of the present invention provides a ball mill load measuring device, including: the signal acquisition module is used for acquiring vibration signal data and audio signal data of the ball mill and collecting the vibration signal data and the audio signal data; the characteristic extraction module is used for extracting the characteristics of the vibration signal data and the audio signal data to obtain vibration IMF component characteristic data and audio mfcc characteristic data; the load measurement module is used for acquiring a pre-trained multi-modal neural network model, and inputting vibration IMF component characteristic data and audio mfcc characteristic data into the multi-modal neural network model to obtain a ball mill load measurement result; the pre-trained multi-mode neural network model comprises a feature extraction network and a two-layer fully-connected BP neural network connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
In a third aspect, the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores machine executable instructions capable of being executed by the processor, and the processor executes the machine executable instructions to implement the ball mill load measuring method according to any one of the foregoing embodiments.
In a fourth aspect, the present invention provides a machine-readable storage medium having stored thereon machine-executable instructions which, when invoked and executed by a processor, cause the processor to implement the ball mill load measurement method of any one of the preceding embodiments.
The invention provides a ball mill load measuring method, a device, electronic equipment and a medium, the method firstly obtains vibration signal data and audio signal data of a ball mill, then performs characteristic extraction on the vibration signal data and the audio signal data to obtain vibration IMF component characteristic data and audio mfcc characteristic data, further obtains a pre-trained multi-mode neural network model, inputs the vibration IMF component characteristic data and the audio mfcc characteristic data into the multi-mode neural network model (the pre-trained multi-mode neural network model comprises a characteristic extraction network and two layers of fully-connected BP neural networks connected with the characteristic extraction network, wherein the characteristic extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network), and obtains a ball mill load measuring result. In the mode, the vibration signal and the audio signal of the ball mill are obtained, and the characteristic extraction is carried out on the vibration signal and the audio signal, so that the useful information of the obtained original signal (namely the vibration signal and the audio signal) is extracted, the accuracy of the prediction model is increased, and the number of training samples is reduced; the method has the advantages that the trained multi-mode neural network model is used for measuring the ball mill load, the filling amount of the ball mill can be accurately and effectively measured, the possibility is provided for real-time control of the filling amount of the ball mill, the working efficiency of the ball mill can be improved, and the power consumption and the raw material consumption are reduced.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic flow chart of a method for measuring a load of a ball mill according to an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a neural network according to an embodiment of the present invention;
FIG. 3 is a schematic flow chart of a specific method for measuring a load of a ball mill according to an embodiment of the present invention;
FIG. 4 is a schematic structural diagram of a ball mill load measuring device according to an embodiment of the present invention;
fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It should be noted that: like reference numbers and letters refer to like items in the following figures, and thus, once an item is defined in one figure, it need not be further defined and explained in subsequent figures.
Some embodiments of the invention are described in detail below with reference to the accompanying drawings. The embodiments described below and the features of the embodiments can be combined with each other without conflict.
If the load of the mill or the filling amount in the mill can be kept to be a certain value or a certain range, the output per machine hour of the mill can be highest, the unit power consumption is lowest, the energy-saving effect is optimal, the mill can stably run for a long time under the optimal working condition, and idle milling and full milling are prevented. Aiming at the problems existing in the current mill load measurement, the embodiment adopts a vibration sensor, an audio sensor and a DSP chip of advanced signal processing to perform algorithm extraction and analysis on multi-sensor data, and a cloud server is used for training a model and storing historical data. The embodiment adopts a method of one set of equipment and one model, and effectively solves the problem of inconsistent results caused by inconsistent working conditions of different environmental noises. The core part is based on a multi-mode deep learning model, and multi-mode fusion is responsible for combining information of multiple modes (in the embodiment, two modes of vibration signals and audio signals) to predict the ball mill load measurement result.
For convenience of understanding, a detailed description will be first given of a method for measuring a load of a ball mill according to an embodiment of the present invention, referring to a schematic flow chart of the method for measuring a load of a ball mill shown in fig. 1, the method mainly includes the following steps S102 to S106:
and step S102, acquiring vibration signal data and audio signal data of the ball mill.
The obtained vibration signal data and the audio signal data of the ball mill can be collected through the collecting device and obtained in a data uploading mode through the collecting device. Wherein, the collection device can include a vibration collection device and an audio collection device.
And step S104, performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data.
The feature extraction is a preprocessing operation performed on the acquired vibration signal data and audio signal data, and in an actual application, the feature extraction processing may be performed by an edge-side device (such as a DSP signal processing chip).
The physical meaning of the mfcc is a group of feature vectors obtained by coding the speech physical information (spectrum envelope and details) in the speech recognition field.
And S106, acquiring a pre-trained multi-modal neural network model, and inputting vibration IMF component characteristic data and audio mfcc characteristic data into the multi-modal neural network model to obtain a ball mill load measurement result.
The multi-modal neural network model is obtained by pre-training in the cloud (also can be called a server or a cloud service center), such as a network model obtained by improving a convolutional neural network CNN, and the input of the multi-modal neural network model is data of multiple modalities. The pre-trained multi-mode neural network model comprises a feature extraction network and a two-layer fully-connected BP neural network connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
According to the ball mill load measuring method provided by the embodiment of the invention, the vibration signal and the audio signal of the ball mill are obtained, the characteristics of the vibration signal and the audio signal are extracted, and the useful information of the original signal (namely the vibration signal and the audio signal) is extracted, so that the accuracy of a prediction model is increased, and the number of training samples is reduced; the method has the advantages that the trained multi-mode neural network model is used for measuring the ball mill load, the filling amount of the ball mill can be accurately and effectively measured, the possibility is provided for real-time control of the filling amount of the ball mill, the working efficiency of the ball mill can be improved, and the power consumption and the raw material consumption are reduced.
In one embodiment, the present embodiment performs multi-sensor information fusion by using late-stage feature fusion because the correlation between the original vibration signal and the audio signal is weak. In addition, the feature fusion mode of the vibration signal and the audio signal can be changed from the current double-layer fully-connected neural network to other modes, such as lstm prediction or linear regression after extracting the principal component. The refinement of the network part can be adjusted according to the situation, for example, the proportion of dropout is added or changed, an attention architecture is added, and the regularization of a loss function is modified.
In consideration of the fact that all sensor data are transmitted back to the cloud server to perform load calculation and machine learning robustness, and the stability is poor, in the embodiment, model training is performed at the cloud end, the trained model (namely, the neural network) is sent to edge equipment (namely, a DSP signal processing chip), and a deep learning algorithm is applied to the edge equipment to measure the load of the ball mill. In an embodiment, the step of training the pre-trained multi-modal neural network model in the cloud may include the following steps 2.1) to 2.4):
step 2.1), obtaining label data and characteristic data; the characteristic data comprises vibration IMF component characteristic data and audio mfcc characteristic data; the obtained label data is given label data after manual labeling.
Step 2.2), performing data enhancement processing on the characteristic data, and performing data truncation processing on the vibration IMF component characteristic data and the audio mfcc characteristic data after the data enhancement processing to obtain a plurality of truncated data;
step 2.3), adding each truncated data serving as a sample into training sample data, and adding white noise data into the training sample data to perform data expansion processing to obtain target training sample data;
and 2.4) training the neural network model by adopting target training sample data.
For step 2.4), in practical application, the target training sample data may be divided into two parts, such as 75% of the target training sample data may be used for training, 25% may be used for verification, and 75% of the training data may be added to the training model for training. And (3) carrying out result verification on the training data and the verification data, and judging the model effect according to the root mean square error of the training data and the verification data, wherein the root mean square error calculation formula is as follows:
Figure M_210909172739649_649296001
the pre-trained multi-modal neural network model comprises a feature extraction network and two layers of fully-connected BP neural networks which are connected in sequence, and a structural schematic diagram of a neural network shown in FIG. 2 is shown. Specifically, the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network; the vibration signal CNN sub-network comprises a Batch Normalization layer (BN), a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer and a Batch Normalization layer which are connected in sequence; the audio signal CNN sub-network comprises a batch standardization layer, an 11 × 11 convolution layer, a pooling layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, a full connection layer and a batch standardization layer which are connected in sequence. Wherein the first layer of batch normalization inputs of the sub-network of vibration signals CNN is a vibration IMF matrix (i.e., vibration IMF component feature data), and the first layer of batch normalization inputs of the sub-network of audio signals CNN is an audio mfcc matrix (i.e., audio mfcc feature data). The last batch normalization layer of the vibration signal CNN sub-network and the last batch normalization layer of the audio signal CNN sub-network are respectively connected with two layers of fully-connected BP neural networks which are sequentially connected, and output is carried out after the second layer of fully-connected BP neural networks.
The above-mentioned vibration signal data and audio signal data of obtaining the ball mill, when being implemented specifically, may include the following steps:
and 3.1) acquiring vibration signal data and audio signal data of the ball mill acquired by the acquisition device based on a preset acquisition frequency to obtain initial acquisition data.
Step 3.2), sampling and band-pass filtering the initial acquisition data; wherein the first filtering range for the vibration signal data and the second filtering range for the audio signal data are different.
Aiming at the step 3.1), the acquisition device comprises a vibration acquisition device and an audio acquisition device, and the acquisition frequency is kept consistent in practical application, for example, one acquisition in 5 minutes.
And 3.2) sampling the vibration signal according to the given sampling frequency, and performing band-pass filtering on the signal, wherein the first filtering range is 12000Hz, the audio signal is adopted according to the given adopted frequency, and the signal is subjected to band-pass filtering, and the second filtering range is 4000 Hz.
In addition, the temperature and current data may be filtered to preserve the data at the closest point, thereby removing data instability and sampling errors.
Further, when the deep learning algorithm is applied to the edge side device (i.e., the DSP signal processing chip) to measure the ball mill load, firstly, feature extraction may be performed on the vibration signal data and the audio signal data based on the DSP signal processing chip to obtain vibration IMF component feature data and audio mfcc feature data. In specific implementation, a preset number of empirical mode decompositions can be performed on vibration signal data based on the DSP signal processing chip to obtain vibration IMF component characteristic data, and meanwhile, mfcc characteristic extraction can be performed on audio signal data based on the DSP signal processing chip to obtain audio mfcc characteristic data.
For the sake of understanding, different processing methods are respectively adopted for the vibration signal data and the audio signal data, and the following detailed description is made:
for vibration signal data, when performing a preset number of empirical mode decompositions on the vibration signal data based on a DSP signal processing chip, the following steps may be adopted:
step S1, determining all local maximum value points and local minimum value points of the vibration signal data, and fitting by a cubic spline interpolation function to form an upper envelope line and a lower envelope line of the original data;
step S2, calculating the mean of the upper envelope and the lower envelope
Figure F_210909172731972_972063014
Determining a new data sequence
Figure F_210909172732114_114628015
(ii) a New data sequence
Figure F_210909172732270_270843016
To remove sequences of low frequencies;
step S3, judging new data sequence
Figure F_210909172732433_433437017
Whether the condition that IMF is established is satisfied, and if so, determining
Figure F_210909172732576_576989018
Repeating the steps S1 and S2 if the first IMF component of the vibration signal data is not satisfied, obtaining
Figure F_210909172732764_764478019
(ii) a Wherein the content of the first and second substances,
Figure M_210909172739728_728907001
a new data sequence confirmed after the first iteration;
Figure M_210909172739776_776314002
the mean value of the upper envelope line and the lower envelope line obtained after the first repeated operation;
step S4, judgment
Figure F_210909172732954_954441020
If the IMF is satisfied, repeating the steps S1 to S3 k times until the IMF is satisfied
Figure F_210909172733114_114228021
，
Figure F_210909172733288_288453022
Finally, the IMF condition is satisfied, and the IMF condition is the first IMF component and is recorded as:
Figure F_210909172733507_507675023
(ii) a Wherein the content of the first and second substances,
Figure M_210909172739822_822598001
a new data sequence confirmed after the kth repeat operation;
Figure M_210909172739854_854376002
a new data sequence confirmed after the k-1 repeated operation;
Figure M_210909172739886_886593003
the mean value of the upper envelope line and the lower envelope line obtained after the kth repeated operation;
Figure M_210909172739917_917896004
is the first IMF component;
step S5, subtracting the vibration signal data
Figure F_210909172733681_681440024
Obtaining a first residual component
Figure F_210909172733869_869513025
Will be
Figure F_210909172734092_092177026
As the signal to be decomposed, the above steps S1 to S4 are performed to obtain the second IMF component
Figure M_210909172739964_964762001
And determining a second residual component
Figure M_210909172740027_027526002
Repeating the above steps S1 to S4 until the nth IMF component is obtained
Figure M_210909172740092_092208003
And an nth residual component to obtain n IMF components after decomposing the vibration signal data
Figure M_210909172740123_123412004
And n residual components
Figure M_210909172740154_154667005
。
For the audio signal data, when performing the mfcc feature extraction on the audio signal data based on the DSP signal processing chip, the following steps 4.1) to 4.3) may be adopted:
step 4.1), performing framing processing on the audio signal data to obtain a plurality of frame data; frame data between two adjacent frames contain repeated data;
step 4.2), carrying out digital signal processing on the frame data of each frame; the digital signal processing comprises adding a Hamming window, Fourier transformation, calculating a power spectrum and triangular band-pass filtering;
and 4.3) calculating the logarithmic energy output by the filter according to the frame data after the digital signal processing, and performing discrete cosine transform to obtain audio mfcc characteristic data.
For the above step 4.1), in order to obtain the local detail information of the signal, the present embodiment omits the conventional pre-emphasis processing step, and directly performs the framing processing on the audio signal data x (t). In order to take into account continuity between frames, successive repeated data is required between frames. In one embodiment, a frame size of 256 and an overlap area of 85 may be taken.
Further, inputting the determined audio mfcc characteristic data and the vibration IMF component characteristic data into a trained neural network model (such as a convolutional neural network) to obtain a final measurement result, uploading the result to a cloud, and storing and displaying a load result by the cloud.
All the parts extracted by the characteristics are spliced together and accessed into a two-layer fully-connected BP neural network to output a load calculation result. The cloud (namely the cloud data service center) comprises a database, a data enhancement module, a vibration signal IMF extraction module, an audio signal mfcc extraction module, a model training module, a model evaluation module and a model issuing management module. The database includes the storage of historical vibration data, audio data and load data, and model management for managing the relationship of the model to the equipment. The models correspond to the equipment, and one equipment corresponds to one set of models; and the data enhancement module is used for intercepting the training data, adding noise, fitting new samples and other contents. The model training module is used for training the model to give a model training result, and the evaluation module is used for evaluating, verifying and storing the model; the model issuing module is used for issuing the model corresponding to the equipment to the edge equipment.
For the convenience of understanding, the present embodiment also provides a specific method for measuring the load of the ball mill, which specifically includes the following steps:
the method comprises the following steps: a standard vibration sensor is arranged on a bearing of the ball mill, a microphone ear sensor is arranged at the center above the ball mill, a current and temperature sensor is arranged, working condition data required by other relevant positions (the working condition data can be not configured) is arranged, and the sampling frequency, the sampling interval and the number of sampling points of the sensor are controlled and collected on a server; and the acquisition sensor acquires data according to the requirements of the server and sends the data to the edge equipment.
Step two: configuring a collecting device corresponding to the edge equipment and configuring a deep learning model corresponding to the edge equipment; the edge device preprocesses the received signal, extracts IMF component from the vibration signal, extracts the mfcc characteristic from the audio signal, transmits the processed characteristic into the deep learning model, gives the result and sends the result to the server.
Step three: and configuring a model updating period, a training data acquisition mode and a setting label at the server side.
Step four: the filtering mode of the vibration sensor and the audio sensor adopts a finite impulse response filter to respectively carry out band-pass filtering on the vibration and the vibration sound signals.
Wherein, when filtering, use
Figure F_210909172734280_280611027
(ii) a In the formula:
Figure F_210909172734487_487152028
is the filter coefficient;
Figure F_210909172734643_643887029
is the output of the filter at point m;
Figure F_210909172734770_770850030
the value of the m-n points of the original data;
Figure F_210909172734928_928705031
n =0,1,2,. for a sampling instant, N-1; the vibration signal adopts a band-pass filter with the frequency range of 100-12000 Hz; the vibration sound signal adopts a band-pass filter of 500-4000 Hz.
Step five: the component extraction of the IMF of the vibration signal is determined as 10 components, first for training the data alignment of the samples and second for sufficient extraction of the IMF component.
Step six: the audio signal mfcc feature extraction order is 40.
Step seven: and optimizing parameters corresponding to the vibration feature extraction part and the audio feature extraction part by adopting a forward transmission mode and a backward transmission mode based on the CNN measurement model parameter optimization.
Further, this embodiment also provides a specific implementation manner, referring to fig. 3, first, a ball mill is given, then, data (vibration and audio) are collected, band-pass filtering is performed on the vibration data and the audio data, Empirical Mode Decomposition (EMD) is performed on the vibration data after the band-pass filtering, an mfcc feature is extracted from the audio data after the band-pass filtering, then, the data are combined after the data normalization processing, and a prediction result is finally given according to feature extraction parameters learned by a model. Aiming at the model, firstly, initializing a label value according to the general CNN deep learning model, verifying whether the label is correct or not, if the current label is incorrect, marking the label value on the data again, and further establishing an independent deep learning CNN model for each ball mill (such as using the general model parameters as initialization and reducing training time); and if the verification label is correct, directly establishing a separate deep learning CNN model for each ball mill, and finally storing the model in mongodb. After the data are combined after the data normalization processing, the data are transmitted to a server through a udp protocol (because the model is trained in the server, the transmitted data are sent to the server so that the server can label the data with a label value), model parameters are downloaded through the udp protocol by the model, so that parameters are extracted according to the characteristics learned by the model to obtain a prediction result, the prediction result is uploaded through the udp protocol, and the prediction result data are stored in mongodb.
Through the above manner, the present embodiment can achieve the following effects:
firstly, the method comprises the following steps: at present, a data acquisition device is adopted for load measurement of a common ball mill, locally acquired data are sent to a cloud center, a prediction result is obtained through calculation and analysis, however, as the number of sensors is increased, the number of the acquisition devices is increased, real-time transmission and storage of the data face huge challenges, and once transmission delay occurs, the effect of subsequent automatic control is influenced. The method adopts the CNN convolutional neural network model trained in the cloud center, issues the trained model to the edge side equipment, can measure the load in real time through calculation in the edge side equipment, can update model data when the measurement is deviated, optimizes the model and ensures the accuracy of model prediction.
Secondly, the method comprises the following steps: in the embodiment, the vibration signals and the audio signals of the ball mill are collected, the original data are processed by a preprocessing method, and the useful information of the original signals is extracted, so that the accuracy of a prediction model is increased, and the number of training samples is reduced;
thirdly, the method comprises the following steps: by adopting the deep learning model, the problem that machine learning needs experts to extract features, and the preprocessing and algorithm parameters are controlled, so that the accuracy is low is solved, and meanwhile, the model can deal with the load measurement conditions under different working conditions because the model utilizes working condition information.
Fourthly: because one set of model of one equipment is adopted, the problem of measurement result errors caused by different environments of different equipment is solved.
The embodiment of the invention solves the problem that the load measurement of the same equipment is inaccurate under different noise environments and different working conditions, and is characterized in that the data of multiple sensors are fused, the vibration and audio data of a ball mill are collected at the same time, the sensor data are uniformly transmitted to a DSP processing chip for preprocessing, the DSP chip can download a model from a server end at regular time, the model and the preprocessed data are used for feature extraction, and finally, the measurement result is given by combining with other working condition information.
In view of the above ball mill load measurement, an embodiment of the present invention provides a ball mill load measurement device, referring to a schematic structural diagram of a ball mill load measurement device shown in fig. 4, the device mainly includes the following components:
the signal acquisition module 41 is configured to acquire vibration signal data and audio signal data of the ball mill;
the feature extraction module 42 is configured to perform feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data;
the load measurement module 43 is configured to obtain a multi-modal neural network model trained in advance on the cloud, and input vibration IMF component feature data and audio mfcc feature data into the multi-modal neural network model to obtain a ball mill load measurement result; the pre-trained multi-mode neural network model comprises a feature extraction network and a two-layer fully-connected BP neural network connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
According to the ball mill load measuring device provided by the embodiment of the invention, the vibration signal and the audio signal of the ball mill are collected, the original data are processed by a digital signal processing method, and the useful information of the original signal is extracted, so that the accuracy of a prediction model is increased, and the number of training samples is reduced; the load measurement is carried out through the neural network model trained by the cloud, the filling amount of the ball mill can be accurately and effectively measured, the possibility is provided for the real-time control of the filling amount of the ball mill, the working efficiency of the ball mill can be improved, and the power consumption and the raw material consumption are reduced.
In one embodiment, the pre-trained multi-modal neural network model is trained in the cloud, and the device further comprises a model training module for acquiring tag data and feature data; the characteristic data comprises vibration IMF component characteristic data and audio mfcc characteristic data; performing data enhancement processing on the characteristic data, and performing data truncation processing on the vibration IMF component characteristic data and the audio mfcc characteristic data after the data enhancement processing to obtain a plurality of truncated data; adding each truncated data serving as a sample into training sample data, and adding white noise data into the training sample data to perform data expansion processing to obtain target training sample data; and training the multi-modal neural network model by adopting target training sample data.
In one embodiment, the sub-network of vibration signals CNN comprises a batch normalization layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, and a batch normalization layer connected in sequence;
the audio signal CNN sub-network comprises a batch standardization layer, an 11 × 11 convolution layer, a pooling layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, a full connection layer and a batch standardization layer which are connected in sequence.
In an embodiment, the signal obtaining module 41 is further configured to obtain vibration signal data and audio signal data of the ball mill, which are collected by the collecting device based on a preset collecting frequency, to obtain initial collecting data; sampling and band-pass filtering the initial acquisition data; wherein the first filtering range for the vibration signal data and the second filtering range for the audio signal data are different.
In an embodiment, the feature extraction module 42 is further configured to perform empirical mode decomposition on the vibration signal data by a preset number to obtain vibration IMF component feature data; and performing mfcc feature extraction on the audio signal data to obtain audio mfcc feature data.
In one embodiment, the feature extraction module 42 is further configured to determine all local maximum points and local minimum points of the vibration signal data in step S1, and form an upper envelope and a lower envelope of the original data by fitting a cubic spline interpolation function; step S2, calculating the mean of the upper envelope and the lower envelope
Figure F_210909172735148_148599032
Determining a new data sequence
Figure F_210909172735433_433878033
(ii) a New data sequence
Figure F_210909172735763_763167034
To remove sequences of low frequencies; step S3, judging new data sequence
Figure F_210909172736078_078455035
Whether the condition that IMF is established is satisfied, and if so, determining
Figure F_210909172736365_365114036
Repeating the steps S1 and S2 if the first IMF component of the vibration signal data is not satisfied, obtaining
Figure F_210909172736712_712428037
(ii) a Wherein the content of the first and second substances,
Figure M_210909172740217_217238001
a new data sequence confirmed after the first iteration;
Figure M_210909172740248_248410002
the mean value of the upper envelope line and the lower envelope line obtained after the first repeated operation; step S4, judgment
Figure F_210909172737042_042369038
Whether or not to satisfy the condition of IMF establishmentIf not, repeating the above steps S1 to S3 k times until obtaining
Figure F_210909172737358_358262039
，
Figure F_210909172737675_675107040
Finally, the IMF condition is satisfied, and the IMF condition is the first IMF component and is recorded as:
Figure F_210909172737990_990108041
(ii) a Wherein the content of the first and second substances,
Figure M_210909172740297_297340003
a new data sequence confirmed after the kth repeat operation;
Figure M_210909172740344_344114004
a new data sequence confirmed after the k-1 repeated operation;
Figure M_210909172740375_375363005
the mean value of the upper envelope line and the lower envelope line obtained after the kth repeated operation;
Figure M_210909172740406_406593006
is the first IMF component; step S5, subtracting the vibration signal data
Figure F_210909172738258_258773042
Obtaining a first residual component
Figure F_210909172738499_499375043
Will be
Figure F_210909172738766_766469044
As the signal to be decomposed, the above steps S1 to S4 are performed to obtain the second IMF component
Figure M_210909172740453_453507007
And determining a second residual component
Figure M_210909172740485_485649008
Repeating the above steps S1 to S4 until the nth IMF component is obtained
Figure M_210909172740533_533077009
And an nth residual component to obtain n IMF components after decomposing the vibration signal data
Figure M_210909172740564_564319010
And n residual components
Figure M_210909172740611_611195011
。
In one embodiment, the feature extraction module 42 is further configured to perform digital signal processing on the frame data of each frame; the digital signal processing comprises adding a Hamming window, Fourier transformation, calculating a power spectrum and triangular band-pass filtering; and calculating the logarithmic energy output by the filter according to the frame data after the digital signal processing, and performing discrete cosine transform to obtain audio mfcc characteristic data.
The device provided by the embodiment of the present invention has the same implementation principle and technical effect as the method embodiments, and for the sake of brief description, reference may be made to the corresponding contents in the method embodiments without reference to the device embodiments.
The embodiment of the invention provides electronic equipment, which particularly comprises a processor and a storage device; the storage means has stored thereon a computer program which, when executed by the processor, performs the method of any of the above described embodiments.
Fig. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention, where the electronic device 100 includes: the device comprises a processor 50, a memory 51, a bus 52 and a communication interface 53, wherein the processor 50, the communication interface 53 and the memory 51 are connected through the bus 52; the processor 50 is arranged to execute executable modules, such as computer programs, stored in the memory 51.
The Memory 51 may include a high-speed Random Access Memory (RAM) and may also include a non-volatile Memory (non-volatile Memory), such as at least one disk Memory. The communication connection between the network element of the system and at least one other network element is realized through at least one communication interface 53 (which may be wired or wireless), and the internet, a wide area network, a local network, a metropolitan area network, and the like can be used.
The bus 52 may be an ISA bus, PCI bus, EISA bus, or the like. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, only one double-headed arrow is shown in FIG. 5, but this does not indicate only one bus or one type of bus.
The memory 51 is used for storing a program, the processor 50 executes the program after receiving an execution instruction, and the method executed by the apparatus defined by the flow process disclosed in any of the foregoing embodiments of the present invention may be applied to the processor 50, or implemented by the processor 50.
The processor 50 may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method may be performed by integrated logic circuits of hardware or instructions in the form of software in the processor 50. The Processor 50 may be a general-purpose Processor, and includes a Central Processing Unit (CPU), a Network Processor (NP), and the like; the device can also be a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA), or other Programmable logic devices, discrete Gate or transistor logic devices, discrete hardware components. The various methods, steps and logic blocks disclosed in the embodiments of the present invention may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present invention may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in the memory 51, and the processor 50 reads the information in the memory 51 and completes the steps of the method in combination with the hardware thereof.
The ball mill load measuring method, device, electronic device and computer program product of the medium provided in the embodiments of the present invention include a computer readable storage medium storing a nonvolatile program code executable by a processor, where the computer readable storage medium stores a computer program, and when the computer program is executed by the processor, the method described in the foregoing method embodiments is executed.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the system described above may refer to the corresponding process in the foregoing embodiments, and is not described herein again.
The computer program product of the readable storage medium provided in the embodiment of the present invention includes a computer readable storage medium storing a program code, where instructions included in the program code may be used to execute the method described in the foregoing method embodiment, and specific implementation may refer to the method embodiment, which is not described herein again.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk or an optical disk, and other various media capable of storing program codes.
Finally, it should be noted that: the above embodiments are only used to illustrate the technical solution of the present invention, and not to limit the same; while the invention has been described in detail and with reference to the foregoing embodiments, it will be understood by those skilled in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some or all of the technical features may be equivalently replaced; and the modifications or the substitutions do not make the essence of the corresponding technical solutions depart from the scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method of measuring load on a ball mill, the method comprising:
acquiring vibration signal data and audio signal data of the ball mill;
performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data;
obtaining a pre-trained multi-modal neural network model, and inputting the vibration IMF component characteristic data and the audio mfcc characteristic data into the multi-modal neural network model to obtain a ball mill load measurement result; the pre-trained multi-mode neural network model comprises a feature extraction network and two layers of fully-connected BP neural networks connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
2. The ball mill load measuring method according to claim 1, wherein the pre-trained multi-modal neural network model is trained in a cloud, and the training step of the pre-trained multi-modal neural network model comprises:
acquiring label data and characteristic data; the feature data comprises vibration IMF component feature data and audio mfcc feature data;
performing data enhancement processing on the characteristic data, and performing data truncation processing on the vibration IMF component characteristic data and the audio mfcc characteristic data after the data enhancement processing to obtain a plurality of truncated data;
adding each truncated data serving as a sample into training sample data, and adding white noise data into the training sample data to perform data expansion processing to obtain target training sample data;
and training a multi-modal neural network model by adopting the target training sample data.
3. The ball mill load measuring method according to claim 1, characterized in that the vibration signal CNN sub-network comprises a batch normalization layer, a 5 x 5 convolution layer, a pooling layer, a 3 x 3 convolution layer, a pooling layer, and a batch normalization layer connected in sequence;
the audio signal CNN sub-network comprises a batch standardization layer, an 11 × 11 convolution layer, a pooling layer, a 5 × 5 convolution layer, a pooling layer, a 3 × 3 convolution layer, a pooling layer, a full connection layer and a batch standardization layer which are connected in sequence.
4. A ball mill load measuring method according to claim 1, wherein said step of acquiring vibration signal data and audio signal data of the ball mill includes:
acquiring vibration signal data and audio signal data of the ball mill acquired by an acquisition device based on a preset acquisition frequency to obtain initial acquisition data;
sampling and band-pass filtering the initial acquisition data; wherein the first filtering range for the vibration signal data and the second filtering range for the audio signal data are different.
5. The ball mill load measuring method according to claim 1, wherein the step of performing feature extraction on the vibration signal data and the audio signal data to obtain vibration IMF component feature data and audio mfcc feature data includes:
performing empirical mode decomposition of a preset number on the vibration signal data to obtain vibration IMF component characteristic data;
and performing mfcc characteristic extraction on the audio signal data to obtain audio mfcc characteristic data.
6. A ball mill load measuring method according to claim 5, wherein the step of performing a preset number of empirical mode decompositions on the vibration signal data to obtain the vibration IMF component characteristic data comprises:
step S1, determining all local maximum value points and local minimum value points of the vibration signal data, and fitting by a cubic spline interpolation function to form an upper envelope line and a lower envelope line of the original data;
step S2, calculating the mean value of the upper envelope line and the lower envelope line
Figure F_210909172723122_122869001
Determining a new data sequence
Figure F_210909172723296_296719002
(ii) a The new data sequence
Figure F_210909172723406_406116003
To remove sequences of low frequencies;
step S3, judging new data sequence
Figure F_210909172723551_551156004
Whether the condition that IMF is established is satisfied, and if so, determining
Figure F_210909172723881_881242005
Repeating the steps S1 and S2 if the first IMF component of the vibration signal data is not satisfied, obtaining
Figure F_210909172724023_023775006
(ii) a Wherein the content of the first and second substances,
Figure M_210909172725784_784089001
a new data sequence confirmed after the first iteration;
Figure M_210909172725846_846560002
the mean value of the upper envelope line and the lower envelope line obtained after the first repeated operation;
step S4, judgment
Figure F_210909172724182_182024007
If the IMF is satisfied, repeating the steps S1 to S3 k times until the IMF is satisfied
Figure F_210909172724357_357780008
，
Figure F_210909172724548_548243009
Finally, the IMF condition is satisfied, and the IMF condition is the first IMF component and is recorded as:
Figure F_210909172724737_737697010
(ii) a Wherein the content of the first and second substances,
Figure M_210909172725911_911036001
a new data sequence confirmed after the kth repeat operation;
Figure M_210909172725957_957937002
a new data sequence confirmed after the k-1 repeated operation;
Figure M_210909172726004_004869003
the mean value of the upper envelope line and the lower envelope line obtained after the kth repeated operation;
Figure M_210909172726036_036015004
is the first IMF component;
step S5, from the vibration signal data
Figure M_210909172726067_067245001
Minus
Figure F_210909172724958_958397011
Obtaining a first residual component
Figure F_210909172725227_227423012
Will be
Figure F_210909172725453_453503013
As the signal to be decomposed, the above steps S1 to S4 are performed to obtain the second IMF component
Figure M_210909172726114_114188002
And determining a second residual component
Figure M_210909172726145_145367003
Repeating the above steps S1 to S4 until the nth IMF component is obtained
Figure M_210909172726176_176626004
And an nth residual component to obtain n IMF components after decomposing the vibration signal data
Figure M_210909172726223_223469005
And n residual components
Figure M_210909172726274_274249006
。
7. The ball mill load measuring method according to claim 5, wherein the step of performing mfcc feature extraction on the audio signal data to obtain audio mfcc feature data includes:
performing framing processing on the audio signal data to obtain a plurality of frame data; frame data between two adjacent frames contain repeated data;
performing digital signal processing on frame data of each frame; the digital signal processing comprises adding a Hamming window, Fourier transformation, calculating a power spectrum and triangular band-pass filtering;
and calculating the logarithmic energy output by the filter according to the frame data after the digital signal processing, and performing discrete cosine transform to obtain the audio mfcc characteristic data.
8. A ball mill load measuring device, said device comprising:
the signal acquisition module is used for acquiring vibration signal data and audio signal data of the ball mill;
the characteristic extraction module is used for extracting the characteristics of the vibration signal data and the audio signal data to obtain vibration IMF component characteristic data and audio mfcc characteristic data;
the load measurement module is used for acquiring a pre-trained multi-modal neural network model, and inputting the vibration IMF component characteristic data and the audio mfcc characteristic data into the multi-modal neural network model to obtain a ball mill load measurement result; the pre-trained multi-mode neural network model comprises a feature extraction network and two layers of fully-connected BP neural networks connected with the feature extraction network; the feature extraction network comprises a vibration signal CNN sub-network and an audio signal CNN sub-network.
9. An electronic device comprising a processor and a memory, the memory storing machine executable instructions executable by the processor, the processor executing the machine executable instructions to implement the ball mill load measurement method of any one of claims 1 to 7.
10. A machine-readable storage medium, characterized in that it stores machine-executable instructions which, when invoked and executed by a processor, cause the processor to carry out the ball mill load measurement method of any one of claims 1 to 7.
CN202111065545.7A 2021-09-13 2021-09-13 Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium Active CN113537160B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111065545.7A CN113537160B (en) 2021-09-13 2021-09-13 Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111065545.7A CN113537160B (en) 2021-09-13 2021-09-13 Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium

Publications (2)

Publication Number Publication Date
CN113537160A true CN113537160A (en) 2021-10-22
CN113537160B CN113537160B (en) 2022-01-18

Family

ID=78093128

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111065545.7A Active CN113537160B (en) 2021-09-13 2021-09-13 Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium

Country Status (1)

Country Link
CN (1) CN113537160B (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114077851A (en) * 2021-11-22 2022-02-22 河北工业大学 FSVC-based ball mill working condition identification method
CN116393217A (en) * 2023-02-24 2023-07-07 华能曲阜热电有限公司 Intelligent monitoring method for material level of steel ball coal mill
CN117839819A (en) * 2024-03-07 2024-04-09 太原理工大学 Online multitasking mill load prediction method based on physical information neural network

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101097282A (en) * 2006-06-30 2008-01-02 中国石油天然气集团公司 Optical fiber safety early warning signal identification system
CN101244403A (en) * 2008-03-17 2008-08-20 西安艾贝尔科技发展有限公司 Optimization control method for grind grading process
CN101871733A (en) * 2010-06-11 2010-10-27 昆明理工大学 Safety detecting method for flue gas waste heat recovery power system of industrial furnace
EP2381275A1 (en) * 2010-04-26 2011-10-26 Sick AG Optoelectronic sensor and method for transmitting and receiving light
CN108614548A (en) * 2018-04-03 2018-10-02 北京理工大学 A kind of intelligent failure diagnosis method based on multi-modal fusion deep learning
CN110135058A (en) * 2019-05-14 2019-08-16 北京工业大学 Mill load parameter prediction method based on multi-modal feature subset selection integrated moulding
CN111256814A (en) * 2020-03-13 2020-06-09 天津商业大学 Tower monitoring system and method
CN112016470A (en) * 2020-08-28 2020-12-01 国网福建省电力有限公司电力科学研究院 On-load tap-changer fault identification method based on sound signal and vibration signal
CN113067839A (en) * 2021-06-02 2021-07-02 中国人民解放军国防科技大学 Malicious encrypted flow detection method based on multi-mode neural network

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101097282A (en) * 2006-06-30 2008-01-02 中国石油天然气集团公司 Optical fiber safety early warning signal identification system
CN101244403A (en) * 2008-03-17 2008-08-20 西安艾贝尔科技发展有限公司 Optimization control method for grind grading process
EP2381275A1 (en) * 2010-04-26 2011-10-26 Sick AG Optoelectronic sensor and method for transmitting and receiving light
CN101871733A (en) * 2010-06-11 2010-10-27 昆明理工大学 Safety detecting method for flue gas waste heat recovery power system of industrial furnace
CN108614548A (en) * 2018-04-03 2018-10-02 北京理工大学 A kind of intelligent failure diagnosis method based on multi-modal fusion deep learning
CN110135058A (en) * 2019-05-14 2019-08-16 北京工业大学 Mill load parameter prediction method based on multi-modal feature subset selection integrated moulding
CN111256814A (en) * 2020-03-13 2020-06-09 天津商业大学 Tower monitoring system and method
CN112016470A (en) * 2020-08-28 2020-12-01 国网福建省电力有限公司电力科学研究院 On-load tap-changer fault identification method based on sound signal and vibration signal
CN113067839A (en) * 2021-06-02 2021-07-02 中国人民解放军国防科技大学 Malicious encrypted flow detection method based on multi-mode neural network

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114077851A (en) * 2021-11-22 2022-02-22 河北工业大学 FSVC-based ball mill working condition identification method
CN114077851B (en) * 2021-11-22 2024-04-23 河北工业大学 FSVC-based ball mill working condition identification method
CN116393217A (en) * 2023-02-24 2023-07-07 华能曲阜热电有限公司 Intelligent monitoring method for material level of steel ball coal mill
CN117839819A (en) * 2024-03-07 2024-04-09 太原理工大学 Online multitasking mill load prediction method based on physical information neural network
CN117839819B (en) * 2024-03-07 2024-05-14 太原理工大学 Online multitasking mill load prediction method based on physical information neural network

Also Published As

Publication number Publication date
CN113537160B (en) 2022-01-18

Similar Documents

Publication Publication Date Title
CN113537160B (en) Ball mill load measuring method, ball mill load measuring device, electronic equipment and medium
CN112766549A (en) Air pollutant concentration forecasting method and device and storage medium
CN105632486B (en) Voice awakening method and device of intelligent hardware
KR101734829B1 (en) Voice data recognition method, device and server for distinguishing regional accent
CN110210513B (en) Data classification method and device and terminal equipment
CN110321913B (en) Text recognition method and device
CN110210660B (en) Ultra-short-term wind speed prediction method
CN115987295A (en) Crop monitoring data efficient processing method based on Internet of things
CN115118511B (en) Abnormal flow identification method, device, electronic equipment and storage medium
CN110428835B (en) Voice equipment adjusting method and device, storage medium and voice equipment
CN111178438A (en) ResNet 101-based weather type identification method
CN113409167A (en) Water quality abnormity analysis method and device
CN111340233A (en) Training method and device of machine learning model, and sample processing method and device
CN110874635B (en) Deep neural network model compression method and device
CN110543869A (en) Ball screw service life prediction method and device, computer equipment and storage medium
KR102423282B1 (en) Apparatus and method for recognizing number of measuring intrument
CN116561927A (en) Digital twin-driven small sample rotary machine residual life prediction method and system
CN111353526A (en) Image matching method and device and related equipment
CN112907541B (en) Palm image quality evaluation model construction method and device
CN115274004A (en) Knowledge reuse-based fermentation process thallus concentration prediction method and system
CN113516025A (en) Hyperspectral image processing method, device and medium based on model construction
CN112926724A (en) Grading method and device for yield of injection molding product and electronic equipment
CN112800813B (en) Target identification method and device
CN117892073B (en) Irrigation area water metering system and water metering method
CN116610080B (en) Intelligent production method of leisure chair and control system thereof

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