CN111260125B - Temperature anomaly detection method for rail vehicle component - Google Patents
Temperature anomaly detection method for rail vehicle component Download PDFInfo
- Publication number
- CN111260125B CN111260125B CN202010032787.5A CN202010032787A CN111260125B CN 111260125 B CN111260125 B CN 111260125B CN 202010032787 A CN202010032787 A CN 202010032787A CN 111260125 B CN111260125 B CN 111260125B
- Authority
- CN
- China
- Prior art keywords
- temperature
- detected
- model
- importance
- prediction
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
- G01K1/00—Details of thermometers not specially adapted for particular types of thermometer
- G01K1/14—Supports; Fastening devices; Arrangements for mounting thermometers in particular locations
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01K—MEASURING TEMPERATURE; MEASURING QUANTITY OF HEAT; THERMALLY-SENSITIVE ELEMENTS NOT OTHERWISE PROVIDED FOR
- G01K13/00—Thermometers specially adapted for specific purposes
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Business, Economics & Management (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Strategic Management (AREA)
- Economics (AREA)
- Human Resources & Organizations (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Operations Research (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- General Business, Economics & Management (AREA)
- Electric Propulsion And Braking For Vehicles (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a method for detecting temperature abnormity of a rail vehicle component. And determining one or more components as objects to be detected with abnormal temperature according to the requirements of driving safety. Constructing a lightweight gradient elevator model, and determining the importance of each driving parameter of the vehicle to the temperature prediction of a single object to be detected; determining an input time step of the characteristic required for predicting the temperature based on the characteristic of the temperature rise hysteresis speed; analyzing the interrelation of the same-vehicle components by combining the spatial position distribution and the Pearson coefficient of the components to be detected, and simultaneously predicting the temperatures of a plurality of objects to be detected based on the concept of the same-class components to be detected; constructing a prediction model, and predicting the time sequence signal by using a convolution network to obtain a residual error value of the real temperature and the predicted temperature; and describing the obtained residual value in a two-dimensional space according to the concept of the same type of components, performing abnormity detection on the residual by using an isolated forest method, and early warning on abnormal component temperature in the running process of the rail vehicle.
Description
Technical Field
The invention belongs to the field of rail vehicle abnormity detection, and relates to a temperature abnormity detection method for rail vehicle components.
Background
In recent years, rail vehicles in China are continuously developed, and the safety and reliability of railways are continuously concerned by the whole industry. In the railway field, Beijing university of transportation, China and south university of transportation, southwest university of transportation, Central Islands' Sichuang research institute and the like all research the traffic safety of railway vehicles.
The key parts such as the axle, the bearing, the brake disc and the like are taken as the key parts influencing the driving safety to bear complex excitation and changeable working conditions, so that the temperature of the parts is increased, and the parts are caused to lose efficacy in serious conditions to influence the driving safety. Therefore, the temperature of the key component in operation is predicted and monitored, a real-time temperature model is established, and an early warning decision is set, so that the operation safety of the component can be effectively improved, and early diagnosis and discovery can be realized.
For early failures, during the increasing computer capacity, statistical learning methods are used to study the history of vehicle traffic based on historical data. Similarly, after the deep learning enters the field of computers, the early failure is further researched through the deep learning, the diagnosis efficiency is continuously improved, the effect is continuously improved, and the early diagnosis and early warning of the failure become possible.
The existing data-driven-based method achieves better results in rail vehicle temperature prediction and becomes the most common prediction method. However, such methods still have certain limitations, such as low prediction accuracy and small application range caused by weak data processing capability of the model, and slow positioning and training caused by prediction of a single target. In summary, it is very important to provide a prediction method with high prediction accuracy, wide application range and fast abnormal point positioning.
Disclosure of Invention
Aiming at the problems of low prediction precision and small application range caused by poor data processing capability of the current model, slow positioning and training caused by prediction of a single target and the like in the prior art, the invention provides a method for performing auxiliary analysis on data acquired by a rail vehicle sensor through a lightweight gradient elevator, selecting the input time step of vehicle history data required by temperature prediction, analyzing the correlation of corresponding parts of vehicles in the same train by combining the spatial position distribution of each part to be detected and a Pearson coefficient, and setting up a method for predicting the temperatures of a plurality of parts at the same time.
In order to realize the functions, the technical scheme of the invention specifically comprises the following technical steps:
the first step is as follows: detecting temperature abnormality of one or more of an axle, a bearing and a brake disc as objects to be detected;
the second step is that: sending the detected temperature abnormal signal to a train operation monitoring system;
the third step: and the train operation monitoring system sends out warning information according to the temperature abnormal signal.
Wherein the first step comprises the steps of:
step 1: constructing a lightweight gradient elevator model, carrying out data analysis on data acquired by a rail vehicle sensor, and determining the importance of each driving parameter of the vehicle on the temperature prediction of a single object to be detected;
step 2: distinguishing parts to be detected according to similar measuring points, and determining the input time step of the characteristics required by temperature prediction based on the characteristic of temperature rise lag speed;
and step 3: analyzing the interrelation of the same-vehicle components by combining the spatial position distribution and the Pearson coefficient of the components to be detected, and simultaneously predicting the temperatures of a plurality of objects to be detected based on the concept of the same-class components to be detected;
and 4, step 4: constructing a prediction model, carrying out time sequence processing on the convolutional neural network, and predicting a time sequence signal by using the convolutional network through the expansion of a convolutional layer, the causal convolution and the residual connection;
and 5: predicting the temperature by adopting the network constructed in the step 4 to obtain a residual error value between the real temperature and the predicted temperature;
step 6: drawing the residual value obtained in the step 5 in a two-dimensional space according to the concept of the same type of components, carrying out abnormal detection on the residual by using an isolated forest method, and considering that the object to be detected is abnormal when abnormal data which are not in accordance with normal distribution appear;
the invention is also characterized in that:
the step 1 is implemented according to the following steps:
step 1.1: in the vehicle operation process, acquiring historical data acquired by a rail vehicle sensor, and preprocessing the data by down-sampling, removing null values and filling missing value processing;
step 1.2: constructing a lightweight gradient elevator model, inputting other preprocessed channel data into the lightweight gradient elevator model by taking the temperature measuring points of the parts to be detected as a prediction target, wherein the other channel data comprise a plurality of characteristics such as static load, dynamic load, traction braking load, vehicle speed, shaft speed, motor speed and the like, and sequencing the importance of the characteristics of the selected parts to be detected on the basis;
step 1.3: the sorted data are sequentially arranged from top to bottom according to the importance of the feature variable names, the sum of the importance is equal to 1 after weighted average calculation, the feature variable with the importance greater than F is selected, F is a threshold value set according to experience, the distribution condition of each index in the importance arrangement is determined according to the distribution condition of each index, the sum of the importance of the feature variable with the importance greater than F is ensured to exceed F, F is the ratio of the importance after dimension reduction and reselection, and is generally set to be 0.85-0.9 of the sum of the importance, and the selected feature represents the total amount of all the previous features;
in the step 2, all parts to be detected are distinguished according to the same type of measuring points, the characteristic input time step length required by temperature prediction is determined based on the characteristic of temperature rise lag speed, the point where the temperature T of each part to be detected changes after the speed v starts from 0 is searched, and the data duration needing to be input into the model is obtained after statistics is carried out according to the average duration.
Step 3, analyzing the correlation of the parts on the same vehicle by combining the spatial position distribution and the Pearson coefficient of the parts to be detected, and simultaneously predicting the temperatures of a plurality of objects to be detected based on the concept of the same type of parts to be detected;
in step 4, the time domain convolution network model is constructed as follows:
constructing a whole time domain convolution network, wherein the core of the network is the expansion of convolution layer, the causal convolution and the residual error connection, the expansion convolution is that when d is 1, the expansion convolution is represented as a traditional convolution, when a larger d is used, the top-level output can represent a larger input range, and the output range from the top level can be represented by a larger dThe receiving domain of the convolution network is effectively expanded; causal convolution then takes into account the time sequence problem, requiring the use of x1,x2,…,xtTo predict ytAnd x cannot be usedt+1And data after time t, thereby enabling yt to be closer to the true value; the residual connection is layer jump connection, the purpose of adding the residual connection is to prevent the data from being difficult to fit due to the fact that the number of model layers is too deep, and the model is helped to better fit the data in a layer jump mode;
the expansion of the convolutional layer is performed using the following formula, where d is the expansion coefficient, k is the convolution kernel size, and s-d · i is the past factor:
the causal convolution is performed using the following equation:
residual concatenation was performed using the following formula:
o＝Activation(x+F(x))
data format x before inputm，n，jAfter the model is output, the data format is xm, a and a are the pre-temperature measurement degrees of the model, and the adjustment is carried out according to the part to be detected. (ii) a
Drawing the residual value obtained in the step 5 in a two-dimensional space according to the concept of the same type of components, carrying out abnormal detection on the residual by using an isolated forest method, considering that the object to be detected is abnormal when abnormal data which are not in accordance with normal distribution appear, and carrying out early warning on the object;
compared with the prior art and method, the method has the following advantages:
(1) actual measurement data detected by a sensor which is generally installed on the existing railway vehicle is used as input data of the model, so that the accuracy and the engineering practicability of model establishment are ensured.
(2) The early warning is carried out before the component to be detected breaks down, so that the running safety is guaranteed, and meanwhile, enough time allowance is provided for the operation scheduling of the rail vehicle.
(3) Aiming at the mode that one model is often adopted by each part in the prior art, the invention provides the technical scheme that one model is applied to a plurality of parts with the same model, so that the model is simplified, the calculated amount is reduced, the efficiency is greatly improved, and meanwhile, the precision is also ensured.
(4) According to the characteristic that temperature rise lag time of parts to be detected at different positions is different, input time step length required by temperature prediction of different positions is determined, and prediction accuracy is improved.
(5) In the time domain convolution network provided by the invention, in order to consider the long-time sequence influence, causal convolution and expansion convolution are adopted, and the yt can be closer to a true value through causal derivation and convolution kernel expansion; residual connection is added to prevent the data from being difficult to fit due to too deep training of the model, and the model is helped to better fit the data in a layer skipping mode; the application is an innovation of a convolutional network in time sequence signal processing, long-time memory can be forgotten continuously compared with the recursion characteristic of a cyclic neural network, the network provided by the invention has the unique complete memory of convolution, the problem of long-time sequence processing is better processed than the cyclic neural network, and the prediction capability of a model is improved fully.
(6) Aiming at the main deviation prediction method of the temperature detection of the rail vehicle at present and few abnormal detection methods, the method provides the method for carrying out outlier detection by using an isolated forest, and combines a bearing temperature prediction method based on a time domain convolution network with an outlier detection method based on the isolated forest to realize the temperature abnormal diagnosis based on prediction.
(7) A plurality of key components influencing driving safety are brought into the component to be detected for abnormal diagnosis, so that the temperature abnormal diagnosis of a plurality of components is realized, the service state of each key component is favorably and comprehensively controlled, and an important means is provided for the operation safety, driving scheduling and operation maintenance of the rail vehicle.
Description of the drawings:
FIG. 1 is a map of the spatial positions of components of a rail vehicle;
FIG. 2 is a temperature rise hysteresis velocity diagram of various bearings of a railway vehicle;
FIG. 3 is a temperature trend chart of similar measuring points of a railway vehicle;
FIG. 4 is a diagram of a time domain convolution network model;
FIG. 5 is a graph of temperature prediction accuracy;
FIG. 6 is a distribution diagram of outliers of an isolated forest;
FIG. 7 is a temperature trend chart of a fault bearing on the day of fault
FIG. 8 is a graph of normal bearing temperature trend during fault day
FIG. 9A flow chart of temperature anomaly detection for a rail vehicle component
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 described clearly and completely with reference to the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present invention, and not all embodiments.
Thus, the following detailed description of the embodiments of the invention is not intended to limit the scope of the invention as claimed, but is merely representative of some 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 the embodiments of the present invention and the features and technical solutions thereof may be combined with each other without conflict.
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.
In the description of the present invention, it should be noted that the terms "upper", "lower", and the like refer to orientations or positional relationships based on those shown in the drawings, or orientations or positional relationships that are conventionally arranged when the products of the present invention are used, or orientations or positional relationships that are conventionally understood by those skilled in the art, and such terms are used for convenience of description and simplification of the description, and do not refer to or imply that the devices or elements referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus, should not be construed as limiting the present invention. Furthermore, the terms "first," "second," and the like are used merely to distinguish one description from another, and are not to be construed as indicating or implying relative importance.
The flow of detecting a temperature abnormality of a rail vehicle component employed in the present invention is shown in fig. 9. For simplicity of description, the invention mainly takes the bearing as a main part and describes the whole process from sensor data processing to temperature abnormity diagnosis of a single component to be detected.
The step 1 is implemented according to the following steps:
step 1.1: in the vehicle operation process, historical data acquired by a rail vehicle sensor is acquired, and the data is preprocessed by down-sampling, null value removal and missing value filling processing.
Step 1.2: constructing a lightweight gradient elevator model, taking bearing temperature measuring points as a prediction target, taking 9 different bearing distribution positions on one shaft as shown in figure 1, inputting other preprocessed channel data into the lightweight gradient elevator model, wherein the other channel data comprises a plurality of characteristics such as static load, dynamic load, traction braking load, vehicle speed, shaft speed, motor speed and the like, and sequencing the importance of the characteristics for the characteristics related to the bearing temperature selected by the invention on the basis. Similarly, for the axle and the brake disc, the axle and brake disc temperature measuring points are taken as prediction targets, the preprocessed other channel data are input into a lightweight gradient elevator model, and feature importance degree sequencing is carried out on the features related to the axle and brake disc temperature selected by the invention on the basis.
The invention mainly uses the algorithm to calculate the feature importance, and obtains the importance arrangement of all features by calculating the importance ratio of the features in each weak learner.
The calculation method of the importance includes two items: gain and frequency. Gain means the relative contribution of a feature to the model calculated by taking the contribution of each feature for each tree in the model, frequency is the number of times the feature is used, and for bearings the importance ranking is shown in table 1.
TABLE 1
Step 1.3: the sorted data are sequentially arranged from top to bottom according to the importance of the feature variable names, the sum of the importance is equal to 1 after weighted average calculation, the feature variable with the importance greater than F is selected, F is a threshold value set according to experience, the distribution condition of each index in the importance arrangement is determined according to the distribution condition of each index, the sum of the importance of the feature variable with the importance greater than F is ensured to exceed F, F is the ratio of the importance after dimension reduction and reselection, and is generally set to be 0.85-0.9 of the sum of the importance, and the selected feature represents the total amount of all the previous features;
because the temperature sensor is arranged in the blind hole of the mounting seat, the temperature rise phenomenon can be generated only when heat is conducted from the heat source to the sensor for a certain time and the calorific value of the bearing is larger than the heat dissipation capacity, so that the temperature rise of the sensor has obvious lag phenomenon relative to the change of the vehicle speed when the vehicle is started;
in the step 2, the bearings are distinguished according to the same type of measuring points, specifically, two types of axle boxes, three types of motors and four types of gear boxes, wherein each type of bearing is four. The gear boxes are divided into a large gear box and a small gear box, so that the four types of the gear boxes are totally four, specifically, two types of axle boxes, three types of motors, two types of large gear boxes and two types of small gear boxes, data analysis is respectively carried out on each type of axle box, points of temperature T change of various bearings after the speed v starts from 0 are searched, and data duration needing to be input into the model is obtained after statistics is carried out according to average duration; as shown in FIG. 2, the lag time of the bearing temperature rise of the four main bearings is different, and the average value is taken to predict in the invention, so that the prediction time of the bearings of the small gear box and the large gear box is 6 minutes, the prediction time of the bearings of the axle box is 13 minutes, and the prediction time of the bearings of the motor is 4 minutes.
The axles and the brake discs are uniformly distributed on the bogie, and the time step lengths of the axles and the brake discs are obtained by the same method.
In the step 3, the bearings are arranged according to the spatial positions of the bearings, the similar bearings are important characteristics for bearing similar working conditions and excitation, and when one bearing is predicted, the other three bearings are input into the model as characteristics, so that the model can be assisted to improve the precision and simultaneously can assist the model to quickly position whether the bearing fails.
According to the design of four axles of two bogies on one axle, nine bearings are distributed on one axle, but not similar bearings, under the condition that other inputs are not changed, only the input is required to be expanded from three similar bearings to twenty-seven similar bearings distributed on different axles, the output is required to be expanded from a single bearing to nine different bearings distributed on one axle, and a single bearing prediction model is converted into a single model to predict nine bearings in total, so that thirty-six prediction models are converted into four prediction models only, and bearing temperature prediction and abnormal detection are accelerated.
For axles, axles of the same type are important characteristics for bearing similar working conditions and excitation, when one axle is predicted, the other three axles are input into the model as the characteristics, so that the model can be helped to improve the accuracy and can be assisted to position whether the temperature of the axle is abnormal or not more quickly.
For the brake disc, the brake discs of the same type are important characteristics for bearing similar working conditions and excitation, when one brake disc is predicted, the brake disc input models at the same positions on the other three axles are used as characteristics, so that the model can be assisted to quickly diagnose whether the brake disc is abnormal in temperature or not while the accuracy of the model is improved.
In step 3, the temperature trends of the similar bearings in fig. 3 can be found to be closer, pearson index analysis is performed on the similar bearings, and the correlation degree of one shaft, three shafts and two and four shafts is the highest in the similar bearings, as shown in table 2;
TABLE 2
For the axle and the brake disc, the same method is adopted to obtain the same type of parts with the highest correlation degree.
In step 4, the time domain convolution network model is constructed as follows:
constructing a whole time domain convolution network, wherein the core of the network is the expansion of convolution layers, the causal convolution and the residual error connection, the expansion convolution is that when d is 1, the expansion convolution is represented as a traditional convolution, and when a larger d is used, the top-level output can represent a larger input range, so that the receiving domain of the convolution network is effectively expanded; causal convolution then takes into account the time sequence problem, requiring the use of x1,x2,…,xtTo predict ytAnd x cannot be usedt+1And data after time t, thereby enabling ytCloser to the true value; the residual connection is layer jump connection, the purpose of adding the residual connection is to prevent the data from being difficult to fit due to the fact that the number of model layers is too deep, and the model is helped to better fit the data in a layer jump mode;
in the constructed network, the dilation convolution, the causal convolution and the residual connection are in a coaction relationship, wherein the dilation convolution and the causal convolution are a type of special operation of convolution, the mode of processing data by a model is improved jointly through dilation and causal deduction, and the residual connection acts on the connection of a network layer, so that the method for processing data by the network layer is not improved;
the expansion of the convolutional layer is performed using the following formula, where d is the expansion coefficient, k is the convolution kernel size, and s-d · i is the past factor:
the causal convolution is performed using the following equation:
residual concatenation was performed using the following formula:
o＝Activation(x+F(x))
data format x before inputm，n，jM represents the size of a batch, n represents a time step, j represents the characteristic quantity of each time step, and after the model is output, the data format is xm，9Namely, the predicted temperature of the multi-output model is obtained;
as shown in fig. 4, for the model structure of the time domain convolution network, the number of convolution kernels is 20, the size is 2, the residual error model is 1, and the dilation convolution is [1,2,4,8,16], that is, the total residual error layers of the model are connected for 5 times;
obtaining a predicted temperature, subtracting the predicted temperature from the actual temperature on the basis to obtain a temperature residual error, as shown in fig. 5, taking a pinion box bearing as an example, and the model verification accuracy provided by the invention is shown in table 3;
TABLE 3
| Mean square error MSE | Root mean square error RMSE | Mean absolute valueFor error MAE | Determining a coefficient R2 |
| 1.40474 | 1.18521 | 0.95543 | 0.99751 |
The same method is used for obtaining temperature residual errors of an axle and a brake disc.
Step 6: according to the correlation analysis shown in table 2, taking a bearing as an example, two similar bearing data sets with the highest correlation are selected for two-dimensional space isolated forest modeling, 200 trees are constructed, the maximum characteristic number is 1, the sampling number is 128, and the data set is obtained by inputting the test set of the time domain convolution model in the step 4;
the isolated forest detection method comprises the following steps:
let T be a node of the orphan tree, which is a leaf node with no children, or only two children (T)l，Tr) The internal node of (2). Each step of segmentation comprises a characteristic q and a segmentation value p, and q is divided<Data of p is divided into TlData with q ≧ p are classified into Tr。
Given n sample data X ═ X1，…，xnD, the dimension of the feature. In order to construct an isolated tree, a feature q and a segmentation value p thereof are randomly selected, and the data set X is recursively segmented until any one of the following conditions is satisfied: (1) the tree has reached a limited height; (2) there is only one sample on a node; (3) all features of the samples on the nodes are the same.
The task of anomaly detection is to give a ranking reflecting the degree of anomaly, and a common ranking method is to rank according to the path length or anomaly score of the sample points, i.e. those points that are ranked first.
Given a data set comprising n samples, the average path length of the tree is
Where H (i) is a harmonic number, which may be estimated as ln (i) + 0.5772156649. c (n) the average of the path lengths for a given number of samples n, to normalize the path length h (x) of the sample x.
The anomaly score for sample x is defined as
Where E (h (x)) is the expected path length of sample x in a collection of isolated trees.
When E (h (x)) → c (n), s → 0.5, i.e., the average path length of the sample x is close to the average path length of the tree, it is not possible to distinguish whether or not there is an abnormality.
When E (h (x) → 0, s → 1, that is, the abnormality score of x approaches 1, it is determined to be abnormal.
When E (h (x)) → n-1, s → 0, it is judged to be normal.
As shown in fig. 6, the abnormal discrimination region constructed by modeling the isolated forest on the test set is shown, and it can be seen that the normal value range of the data can be well demarcated by the demarcation region for finally carrying out the abnormal discrimination by the model;
as shown in fig. 7, the diagnosis result and early warning of a failed bearing of a certain vehicle are shown, and fig. 8 shows the prediction and diagnosis result of a normal bearing of the same vehicle on the same day, it can be seen that the model can correctly diagnose the normal bearing and the failed bearing, and early warning is carried out on the failed bearing 5 hours or more in advance.
The above embodiments are only used for illustrating the invention and not for limiting the technical solutions described in the invention, and although the present invention has been described in detail in the present specification with reference to the above embodiments, the present invention is not limited to the above embodiments, and therefore, any modification or equivalent replacement of the present invention is made; all such modifications and variations are intended to be included herein within the scope of this disclosure and the appended claims.
Claims (6)
1. A method of detecting a temperature anomaly in a rail vehicle component, comprising the steps of:
the first step is as follows: detecting temperature abnormality of one or more of an axle, a bearing and a brake disc as objects to be detected;
the second step is that: sending the detected temperature abnormal signal to a train operation monitoring system;
the third step: the train operation monitoring system sends out warning information according to the temperature abnormal signal;
wherein the first step comprises the steps of;
step 1: constructing a lightweight gradient elevator model, carrying out data analysis on data acquired by a rail vehicle sensor, and determining the importance of each driving parameter of the vehicle on the temperature prediction of a single object to be detected;
step 2: distinguishing parts to be detected according to similar measuring points, and determining the input time step of the characteristics required by temperature prediction based on the characteristic of temperature rise lag speed;
and step 3: analyzing the interrelation of the same-vehicle components by combining the spatial position distribution and the Pearson coefficient of the components to be detected, and simultaneously predicting the temperatures of a plurality of objects to be detected based on the concept of the same-class components to be detected;
and 4, step 4: constructing a prediction model, carrying out time sequence processing on the convolutional neural network, and predicting a time sequence signal by using the convolutional network through the expansion of a convolutional layer, the causal convolution and the residual connection;
and 5: predicting the temperature by adopting the network constructed in the step 4 to obtain a residual error value between the real temperature and the predicted temperature;
step 6: and (5) describing the residual error value obtained in the step (5) in a two-dimensional space according to the concept of the same type of components, carrying out abnormity detection on the residual error by using an isolated forest method, and considering that the object to be detected is abnormal when abnormal data which are not in accordance with normal distribution appear.
2. The method for detecting a temperature abnormality of a railway vehicle component according to claim 1, characterized in that: the step 1 specifically comprises the following steps; step 1.1: in the vehicle operation process, acquiring historical data acquired by a rail vehicle sensor, and preprocessing the data by down-sampling, removing null values and filling missing value processing; step 1.2: constructing a lightweight gradient elevator model, inputting the preprocessed other channel data into the lightweight gradient elevator model by taking the temperature measuring points of the parts to be detected as prediction targets, and sequencing the feature importance of the features related to the parts to be detected on the basis; step 1.3: the sorted data are sequentially arranged from top to bottom according to the importance of the feature variable names, the sum of the importance calculated through weighted average is equal to 1, the feature variables with the importance greater than F are selected, the sum of the importance of the feature variables with the importance greater than F is ensured to exceed F, F is the ratio of the importance after being reselected through dimensionality reduction, the ratio is set to be 0.85-0.9 of the sum of the importance, and the selected features represent F of the total amount of all the previous features.
3. The method for detecting a temperature abnormality of a railway vehicle component according to claim 2, characterized in that: in the step 2, all parts to be detected are distinguished according to the same type of measuring points, the characteristic input time step length required by temperature prediction is determined according to the characteristic of temperature rise lag speed, the point where the temperature T of each part to be detected changes after the speed v starts from 0 is searched, and the data duration needing to be input into the model is obtained after statistics is carried out according to the average duration.
4. The method for detecting a temperature abnormality of a railway vehicle component according to claim 3, characterized in that: in the step 3 of bearing detection, the three input models of the other bearings are used as characteristics when one bearing is predicted according to the arrangement of the spatial positions of the bearings, nine bearings are distributed on one shaft and are not similar to the bearing according to the design of four shafts of two bogies of one vehicle, the three similar bearings are expanded to twenty-seven similar bearings distributed on different shafts according to the condition that other inputs are not changed, the single bearing prediction model is converted into a single model to predict the nine bearings, and then thirty-six prediction models are converted into only four prediction models.
5. The method for detecting a temperature abnormality of a railway vehicle component according to claim 4, characterized in that: in the step 3 of axle detection, similar axles bear important characteristics of similar working conditions and excitation, and when one axle is predicted, the other three axles are input into a model to serve as characteristics.
6. The method for detecting a temperature abnormality of a railway vehicle component according to claim 4, characterized in that: in the step 3 of brake disc detection, the brake discs of the same type are important characteristics for bearing similar working conditions and excitation, and when one brake disc is predicted, the brake discs at the same positions on the other three axles are input into a model to serve as characteristics.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010032787.5A CN111260125B (en) | 2020-01-13 | 2020-01-13 | Temperature anomaly detection method for rail vehicle component |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010032787.5A CN111260125B (en) | 2020-01-13 | 2020-01-13 | Temperature anomaly detection method for rail vehicle component |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111260125A CN111260125A (en) | 2020-06-09 |
| CN111260125B true CN111260125B (en) | 2022-03-01 |
Family
ID=70952416
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010032787.5A Active CN111260125B (en) | 2020-01-13 | 2020-01-13 | Temperature anomaly detection method for rail vehicle component |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111260125B (en) |
Families Citing this family (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111829693B (en) * | 2020-07-30 | 2021-10-22 | 成都运达科技股份有限公司 | Bearing temperature detection method and system based on LSTM model of time and space dimensions |
| CN111815209A (en) * | 2020-09-10 | 2020-10-23 | 上海冰鉴信息科技有限公司 | Data dimension reduction method and device applied to wind control model |
| CN113283003B (en) * | 2020-11-13 | 2022-06-10 | 西南交通大学 | High-speed train axle temperature anomaly detection method based on space-time fusion decision |
| CN112763091B (en) * | 2020-12-30 | 2022-04-15 | 西南交通大学 | Intelligent detection device and test method for temperature signals of subway bolts |
| CN112862219A (en) * | 2021-03-19 | 2021-05-28 | 浙江大学 | Hydroelectric generating set temperature prediction method based on time domain convolution and cyclic neural network |
| CN113190991B (en) * | 2021-04-26 | 2024-04-16 | 联合汽车电子有限公司 | Bearing temperature detection method and system for electric drive axle |
| CN113253125B (en) * | 2021-05-19 | 2023-02-17 | 北方工业大学 | Information fusion-based lithium iron phosphate battery thermal runaway monitoring method and system |
| CN114137587B (en) * | 2021-12-01 | 2022-07-29 | 西南交通大学 | Method, device, equipment and medium for estimating and predicting position of moving object |
| CN117113107B (en) * | 2023-10-24 | 2023-12-29 | 广东顺德爱顺机电科技有限公司 | Rotor die casting abnormal error data processing method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101072985A (en) * | 2004-12-06 | 2007-11-14 | 通用电气公司 | Train wheel bearing temperature detection |
| CN109615121A (en) * | 2018-11-27 | 2019-04-12 | 西安理工大学 | Bullet train axle temperature predicting method based on data-driven support vector machines |
| CN110149258A (en) * | 2019-04-12 | 2019-08-20 | 北京航空航天大学 | A kind of automobile CAN-bus network data method for detecting abnormality based on isolated forest |
| CN110243599A (en) * | 2019-07-02 | 2019-09-17 | 西南交通大学 | Multidimensional peels off train EMU axle box bearing temperature anomaly state monitoring method |
| CN110271581A (en) * | 2019-06-11 | 2019-09-24 | 武汉创牛科技有限公司 | A kind of vehicle trouble acquires maintenance system in real time |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9151681B2 (en) * | 2012-12-19 | 2015-10-06 | Progress Rail Services Corporation | Temperature detector having different types of independent sensors |
-
2020
- 2020-01-13 CN CN202010032787.5A patent/CN111260125B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101072985A (en) * | 2004-12-06 | 2007-11-14 | 通用电气公司 | Train wheel bearing temperature detection |
| CN109615121A (en) * | 2018-11-27 | 2019-04-12 | 西安理工大学 | Bullet train axle temperature predicting method based on data-driven support vector machines |
| CN110149258A (en) * | 2019-04-12 | 2019-08-20 | 北京航空航天大学 | A kind of automobile CAN-bus network data method for detecting abnormality based on isolated forest |
| CN110271581A (en) * | 2019-06-11 | 2019-09-24 | 武汉创牛科技有限公司 | A kind of vehicle trouble acquires maintenance system in real time |
| CN110243599A (en) * | 2019-07-02 | 2019-09-17 | 西南交通大学 | Multidimensional peels off train EMU axle box bearing temperature anomaly state monitoring method |
Non-Patent Citations (1)
| Title |
|---|
| 基于输入限制的I-ELM机车热轴等级判别模型;宋平岗 等;《轴承》;20150731(第428期);第45-49页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111260125A (en) | 2020-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111260125B (en) | Temperature anomaly detection method for rail vehicle component | |
| CN111207938B (en) | Railway vehicle fault detection method | |
| Luo et al. | Convolutional neural network based on attention mechanism and Bi-LSTM for bearing remaining life prediction | |
| KR102226687B1 (en) | Apparatus and method of remaining maintenance cycle prediction based on times series prediction using deep learning | |
| Helmi et al. | Rolling bearing fault detection of electric motor using time domain and frequency domain features extraction and ANFIS | |
| Tsui et al. | Prognostics and health management: A review on data driven approaches | |
| CN112418277B (en) | Method, system, medium and equipment for predicting residual life of rotating machine parts | |
| Reddy et al. | Anomaly detection and fault disambiguation in large flight data: A multi-modal deep auto-encoder approach | |
| Lee et al. | Introduction to data-driven methodologies for prognostics and health management | |
| Li et al. | Multi-sensor data-driven remaining useful life prediction of semi-observable systems | |
| CN113221277A (en) | Bearing performance degradation evaluation method and system based on digital twin model | |
| Yan et al. | A survey on fault diagnosis approaches for rolling bearings of railway vehicles | |
| CN110991471B (en) | Fault diagnosis method for high-speed train traction system | |
| Atamuradov et al. | Feature selection and fault‐severity classification–based machine health assessment methodology for point machine sliding‐chair degradation | |
| Yao et al. | Improving rolling bearing fault diagnosis by DS evidence theory based fusion model | |
| Gao et al. | A process fault diagnosis method using multi‐time scale dynamic feature extraction based on convolutional neural network | |
| Qin et al. | Remaining useful life prediction for rotating machinery based on optimal degradation indicator | |
| Zhang et al. | Remaining Useful Life Prediction of Rolling Bearings Using Electrostatic Monitoring Based on Two‐Stage Information Fusion Stochastic Filtering | |
| CN114356897A (en) | Electromechanical actuator fault diagnosis method based on data fusion | |
| Kannan et al. | Nominal features-based class specific learning model for fault diagnosis in industrial applications | |
| CN112329520A (en) | Truck bearing fault identification method based on generation countermeasure learning | |
| Forest et al. | Large-scale vibration monitoring of aircraft engines from operational data using self-organized Models | |
| Xu et al. | Fault diagnosis of subway traction motor bearing based on information fusion under variable working conditions | |
| Marquez et al. | Digital twins in condition-based maintenance apps: A case study for train axle bearings | |
| Jiang et al. | A SVDD and K‐Means Based Early Warning Method for Dual‐Rotor Equipment under Time‐Varying Operating Conditions |
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 |