Multi-scale instrument fault diagnosis method based on ensemble learning

Abstract

The invention discloses a multi-scale instrument fault diagnosis method based on ensemble learning, and relates to the technical field of fault diagnosis, and the method comprises the following steps: S1, converting multi-scale fault information; s2, acquiring instrument fault diagnosis results of the plurality of models; and S3, fusing a plurality of diagnosis decisions from the fault information of different scales. According to the multi-scale instrument fault diagnosis method based on the ensemble learning, different diagnosis methods are integrated to improve the diagnosability, so that the diagnosis method based on the ensemble learning is tedious in preprocessing and the multi-scale fault sequence contains various fault information; the problems that an integration method for instrument fault diagnosis is tedious in preprocessing and low in diagnosis rate are solved, convenient, practical and accurate instrument fault diagnosis is achieved, and the reliability of industrial application and use of instruments can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a multi-scale instrument fault diagnosis method based on ensemble learning. Background Art

[0002] Instruments are an important part of the information technology industry and play a key role in the acquisition, storage, transmission, processing and utilization of information. Instrument fault diagnosis is not only related to production efficiency, but also directly affects the safety of equipment and personnel, and prevents or eliminates potential catastrophic accidents. Timely instrument fault diagnosis is the key to ensuring the normal operation of the automation system, and can effectively reduce downtime and maintenance costs. The current methods focus on integrating different diagnostic methods to improve diagnostic performance. The preprocessing steps of different diagnostic methods are complicated, making the preprocessing of diagnostic methods based on ensemble learning cumbersome. On the other hand, the current methods do not consider multi-scale integration to improve the performance of instrument fault diagnosis, resulting in insufficient diagnostic performance. In summary, it is necessary to consider multi-scale and ensemble learning for instrument fault diagnosis. Summary of the invention

[0003] The purpose of the present invention is to provide a multi-scale instrument fault diagnosis method based on ensemble learning, so as to solve the problems of cumbersome preprocessing and low diagnosis rate of the ensemble method for instrument fault diagnosis.

[0004] To achieve the above object, the present invention provides a multi-scale instrument fault diagnosis method based on ensemble learning, comprising the following steps:

[0005] Step S1: compress the fault sequence into multi-granularity subsequences using an average pooling layer, so that the multi-scale fault information can be extracted from the subsequences and the original sequence with a fixed kernel size, thereby realizing multi-scale fault information conversion;

[0006] Step S2: input the time series after multi-scale fault information conversion into multiple identical full convolution diagnosis models to obtain instrument fault diagnosis results of multiple models;

[0007] Step S3: fuse multiple diagnostic decisions from fault information of different scales, and calculate again through the fully connected layer to obtain the final instrument failure mode probability distribution.

[0008] Preferably, in step S1, an average pooling operation is used. k-1 Extract multi-scale fault information and perform average pooling k-1 As shown below:

[0009]

[0010] Among them, F kjIndicates that the average pooling operation is performed k-1 To obtain the jth value of the vector, the length of the sub-sampled pooling domain is 1×k matrix and the step size of the average pooling process is set to k, x s Represents the sth value of the input sample x, input H represents the length of the input sample time window.

[0011] Preferably, in step S1, there are n average pooling layers for multi-scale fault information conversion, and n is obtained by the following formula:

[0012]

[0013] Where l represents the length of the time series samples, m is the maximum value of n and is set to 5, and f represents the size of the first filter of the original model.

[0014] Preferably, the specific steps of step S2 include:

[0015] Input multiple time series transformed from scaled fault information into n fully convolutional network diagnosis models, each of which consists of three feature extractors, a global average pooling layer, and a softmax layer;

[0016] The feature extractor of FCN is as follows:

[0017]

[0018] in, BN and ReLU represent the convolution operator, batch normalization layer and rectified linear unit respectively. con represents the convolution kernel parameters of the full convolution layer, b con Represents the deviation of the convolution kernel; the filter sizes of the three convolution operators are 128, 256, and 128, respectively, and the one-dimensional kernel sizes of the three convolution operators are 8, 5, and 3, respectively;

[0019] Each diagnostic model obtains a diagnostic result through the softmax layer.

[0020] Preferably, the specific steps of step S3 include:

[0021] The n diagnostic decisions from fault information of different scales are fused together to form the final fault mode probability distribution;

[0022] Given the i-th diagnosis decision Where c represents the number of fault modes, and the fusion output of n diagnostic decisions is as follows:

[0023]

[0024] The probability of the zth failure mode is obtained as follows:

[0025] v = WF + b;

[0026]

[0027] in, is the network parameter of the fully connected layer, b represents the offset, c represents the number of fault modes, e represents the natural constant, v represents the fault probability vector, and d represents the number of accumulated calculations.

[0028] Therefore, the present invention adopts the above-mentioned multi-scale instrument fault diagnosis method based on ensemble learning, integrates different diagnostic methods to improve diagnostic performance, makes the preprocessing of the diagnostic method based on ensemble learning cumbersome and the multi-scale fault sequence contains diverse fault information, solves the problems of cumbersome preprocessing and low diagnostic rate of the integrated method of instrument fault diagnosis, realizes convenient, practical and accurate instrument fault diagnosis, and is conducive to improving the reliability of industrial application of instruments.

[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 A flowchart of an embodiment of a multi-scale instrument fault diagnosis method based on ensemble learning according to the present invention;

[0031] Figure 2 A schematic diagram of a fully convolutional network diagnostic model of an embodiment of a multi-scale instrument fault diagnosis method based on ensemble learning of the present invention;

[0032] Figure 3 It is a probability density distribution diagram of the accuracy gain of an embodiment of a multi-scale instrument fault diagnosis method based on ensemble learning in the present invention. DETAILED DESCRIPTION

[0033] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0034] Unless otherwise defined, the technical terms or scientific terms used in the present invention should be understood by people with ordinary skills in the field to which the present invention belongs. The words "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0035] Example

[0036] See also Figure 1-3 The present invention provides a multi-scale instrument fault diagnosis method based on ensemble learning, comprising the following steps:

[0037] Step S1: compress the fault sequence into multi-granularity subsequences using an average pooling layer, so that the multi-scale fault information can be extracted from the subsequences and the original sequence with a fixed kernel size, thereby realizing multi-scale fault information conversion;

[0038] The goal of this method is to extract multi-scale fault information while maintaining the ease of use of the multi-scale integrated instrument diagnosis model. It uses an operation similar to the spatial pyramid pooling layer. In order to retain the local fault information of the time series as much as possible, the average pooling operation is used. k-1 Extract multi-scale fault information instead of the maximum pooling operation. In addition, the transformed fault information is sent to multiple original models separately instead of concatenating them into a one-dimensional vector. Averagepooling k-1 As shown below:

[0039]

[0040] Among them, F kj Indicates that the average pooling operation is performed k-1 To obtain the jth value of the vector, the length of the sub-sampled pooling domain is 1×k matrix and the step size of the average pooling process is set to k, x s Represents the sth value of the input sample x, input H represents the length of the input sample time window.

[0041] In the multi-scale fault information conversion stage, there are n average pooling layers for multi-scale fault information conversion. n is the only hyperparameter that needs to be set in this method, and a simple but effective rule is set for it. Set a maximum value for n and try to set n to a larger value within the limits of the integrated model. n is obtained by the following formula:

[0042]

[0043] Where l represents the length of the time series samples, m is the maximum value of n and is set to 5, and f represents the size of the first filter of the original model.

[0044] After the multi-scale fault information conversion stage, the original fault sequence is transformed into a fault sequence of n scales, which enables the original model to have the ability to extract multi-scale fault information and can help extract multi-scale fault feature information without changing the structure and hyperparameters of the integrated model.

[0045] Step S2: Input the time series after multi-scale fault information conversion into multiple identical full convolution diagnosis models to obtain instrument fault diagnosis results of multiple models. The specific steps include:

[0046] The multiple time series transformed by scaled fault information are input into n fully convolutional network diagnosis models. Each fully convolutional network diagnosis model consists of three feature extractors, a global average pooling layer and a softmax layer. The feature extractor of FCN is shown as follows:

[0047]

[0048] in, BN and ReLU represent the convolution operator, batch normalization layer and rectified linear unit respectively. con represents the convolution kernel parameters of the full convolution layer, b con Represents the bias of the convolution kernel. The filter sizes of the three convolution operators are 128, 256, and 128, respectively, and the one-dimensional kernel sizes of the three convolution operators are 8, 5, and 3, respectively.

[0049] Each diagnostic model is Figure 2 As shown, each diagnostic model obtains a diagnostic result through the softmax layer.

[0050] Step S3: fuse multiple diagnostic decisions from fault information of different scales, and calculate again through the fully connected layer to obtain the final instrument failure mode probability distribution. The specific steps include:

[0051] The n diagnostic decisions from fault information of different scales are fused together to form the final fault mode probability distribution.

[0052] Given the i-th diagnosis decision Where c represents the number of failure modes. The fusion output of n diagnostic decisions is as follows:

[0053]

[0054] The probability of the zth failure mode is obtained as follows:

[0055] v = WF + b;

[0056]

[0057] in, is the network parameter of the fully connected layer, b represents the offset, c represents the number of fault modes, e represents the natural constant, v represents the fault probability vector, and d represents the number of accumulated calculations.

[0058] Figure 2 and Figure 3 The probability density distribution of the full convolutional network diagnosis model and accuracy gain provided in this embodiment. Comparing this method with the full convolutional network diagnosis model without considering multi-scale integration can illustrate the applicability and effectiveness of this method for classification results, and that this method has a high diagnostic accuracy.

[0059] Therefore, the present invention adopts the above-mentioned multi-scale instrument fault diagnosis method based on ensemble learning, integrates different diagnostic methods to improve diagnostic performance, makes the preprocessing of the diagnostic method based on ensemble learning cumbersome and the multi-scale fault sequence contains diverse fault information, solves the problems of cumbersome preprocessing and low diagnostic rate of the integrated method of instrument fault diagnosis, realizes convenient, practical and accurate instrument fault diagnosis, and is conducive to improving the reliability of industrial application of instruments.

[0060] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A multi-scale instrument fault diagnosis method based on ensemble learning, characterized in that: The following steps are involved: Step S1: compress the fault sequence into multi-granularity subsequences using an average pooling layer, so that the multi-scale fault information can be extracted from the subsequences and the original sequence with a fixed kernel size, thereby realizing multi-scale fault information conversion; Step S2: input the time series after multi-scale fault information conversion into multiple identical full convolution diagnosis models to obtain instrument fault diagnosis results of multiple models; Step S3: fuse multiple diagnostic decisions from fault information of different scales, and calculate again through the fully connected layer to obtain the final instrument failure mode probability distribution.

2. A multi-scale instrument fault diagnosis method based on ensemble learning according to claim 1, characterized in that: Averagepooling is used in step S1 k-1 Extract multi-scale fault information and perform average pooling k-1 As shown below: Among them, F kj Indicates that the average pooling operation is performed k-1 To obtain the jth value of the vector, the length of the sub-sampled pooling domain is 1×k matrix and the step size of the average pooling process is set to k, x s Represents the sth value of the input sample x, input H represents the length of the input sample time window.

3. The multi-scale instrument fault diagnosis method based on ensemble learning according to claim 2 is characterized in that: In step S1, there are n average pooling layers for multi-scale fault information conversion, and n is obtained by the following formula: Where l represents the length of the time series samples, m is the maximum value of n and is set to 5, and f represents the size of the first filter of the original model.

4. The multi-scale instrument fault diagnosis method based on ensemble learning according to claim 3 is characterized in that: The specific steps of step S2 include: Input multiple time series transformed from scaled fault information into n fully convolutional network diagnosis models, each of which consists of three feature extractors, a global average pooling layer, and a softmax layer; The feature extractor of FCN is as follows: in, BN and ReLU represent the convolution operator, batch normalization layer and rectified linear unit respectively. con represents the convolution kernel parameters of the full convolution layer, b con Represents the deviation of the convolution kernel; the filter sizes of the three convolution operators are 128, 256, and 128, respectively, and the one-dimensional kernel sizes of the three convolution operators are 8, 5, and 3, respectively; Each diagnostic model obtains a diagnostic result through the softmax layer.

5. The multi-scale instrument fault diagnosis method based on ensemble learning according to claim 4 is characterized in that: The specific steps of step S3 include: The n diagnostic decisions from fault information of different scales are fused together to form the final fault mode probability distribution; Given the i-th diagnosis decision Where c represents the number of fault modes, and the fusion output of n diagnostic decisions is as follows: The probability of the zth failure mode is obtained as follows: v = WF + b; in, is the network parameter of the fully connected layer, b represents the offset, c represents the number of fault modes, e represents the natural constant, v represents the fault probability vector, and d represents the number of accumulated calculations.

Images