A method and system for fault diagnosis of industrial equipment based on multi-view generation algorithm

Abstract

This invention discloses a method and system for fault diagnosis of industrial equipment based on a multi-view generation algorithm, belonging to the field of machine learning technology, to solve the problem of low fault diagnosis accuracy caused by inaccurate feature extraction in existing fault diagnosis methods. The key technical points of this invention include: collecting one or more operating state variables of industrial equipment, and sorting the same operating state variables by time to form an original univariate time series set; generating multiple views from multiple time series samples, and combining time series transformation with feature vector concatenation to generate a feature vector set for multiple views; inputting the feature vector of each view into a machine learning-based fault diagnosis classifier for training; generating multiple views from the time series of the operating state sample to be tested to obtain the corresponding feature vector; and inputting the feature vector into the trained classifier to obtain the fault prediction result. This invention significantly improves the accuracy and performance of time series classification and fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, specifically to a method and system for diagnosing industrial equipment faults based on a multi-view generation algorithm. Background Technology

[0002] Industry is a reflection of a nation's strength, and its level is crucial to the national economy and people's livelihood. Industrial production involves a large number of industrial equipment, and the safety and reliability of this equipment are paramount. Equipment malfunctions can not only reduce production efficiency but also cause serious loss of life and property. In modern industry, numerous sensors are installed on industrial equipment to monitor various operational parameters, such as temperature, speed, and pressure. These sensor-collected parameters are arranged into time series, and processing and analyzing these time series enables fault diagnosis.

[0003] Existing industrial equipment fault diagnosis methods employ machine learning-based algorithms, which classify the univariate time series data collected by each sensor to achieve fault diagnosis. These methods offer advantages such as strong interpretability, high speed, and high accuracy, and are widely used.

[0004] However, traditional machine learning-based univariate time series classification methods used for fault diagnosis extract overly simplistic features, failing to fully represent and extract multi-domain fault patterns. A single data description of an object represents one view, while multiple descriptions represent multiple views. Methods that improve classification and diagnostic performance by learning from multiple view data are called multi-view learning. Multi-view learning can extract features from multiple domains, resulting in comprehensive and diverse data descriptions. The descriptions in each view are not entirely identical, exhibiting diversity. By improving the accuracy of each view and the diversity between views, classification accuracy can be enhanced, thereby improving the fault diagnosis performance of industrial equipment. Summary of the Invention

[0005] To address this issue, the present invention proposes an industrial equipment fault diagnosis method and system based on a multi-view generation algorithm, which solves the problem of low fault diagnosis accuracy caused by inaccurate feature extraction in existing machine learning-based fault diagnosis methods.

[0006] According to one aspect of the present invention, a method for fault diagnosis of industrial equipment based on a multi-view generation algorithm is provided, the method comprising the following steps:

[0007] Step 1: Collect one or more operating status variables of industrial equipment, and sort the same type of operating status variables by time to form an original univariate time series set;

[0008] Step 2: Preprocess the original univariate time series set;

[0009] Step 3: Generate multiple views from multiple time series samples in the preprocessed original univariate time series set, and generate feature vector sets for multiple views by combining time series transformation with feature vector concatenation.

[0010] Step 4: Input the feature vector set of each view into the machine learning-based fault diagnosis classifier for training, and obtain the trained fault diagnosis classifier.

[0011] Step 5: Generate multiple views from the time series of the preprocessed test running status samples and obtain the corresponding feature vectors;

[0012] Step 6: Input the feature vector corresponding to the sample of the operating state to be tested into the trained fault diagnosis classifier to obtain the fault prediction result of the sample of the operating state to be tested.

[0013] Furthermore, the operating status quantities mentioned in step one include temperature, rotational speed, and pressure; the industrial equipment includes construction machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment.

[0014] Furthermore, the preprocessing in step two includes: filling in time series with missing values ​​using interpolation methods; processing unequal time series into equal time series using a sliding window method; standardizing each equal time series; and labeling each standardized time series sample.

[0015] Furthermore, the specific process of step three includes:

[0016] Multiple time series samples are subjected to K fast sequence transformations to obtain K transformed time series; the sequence transformations include Hilbert transform, differential transforms of various orders, wavelet transform, or Fourier transform;

[0017] Multiple time series samples and the transformed time series are transformed by random convolution kernels with kernel numbers C1 and C2 respectively, generating corresponding feature vectors. Each time series sample generates K+1 feature vectors. Specifically, each convolution kernel performs a sliding dot product operation on each time series to generate a feature map. P pooling features are extracted from each feature map, wherein the pooling features include the proportion of positive values, the proportion of negative values, the proportion of zero values, the maximum value, the minimum value, and the mean value.

[0018] The feature vectors generated from multiple time series samples and the feature vectors generated from the transformed time series are concatenated in rows to obtain feature vectors for K views. Each feature vector contains (C1+C2)P features.

[0019] Furthermore, in step three, the time series samples s = (s1, s2, ..., s) in the random convolution kernel transformation...n The i-th value s i The formula for the dot product operation with the convolution kernel ω is:

[0020]

[0021] Where * represents the dot product operation, d is the dilation coefficient, l represents the length of the convolution kernel ω, and ω j Let ω represent the j-th value of the convolution kernel ω, and b represent the bias coefficient.

[0022] Furthermore, the specific process in step six includes: inputting the feature vectors of the K views into the corresponding trained fault diagnosis classifiers, and outputting K predicted labels; integrating the K predicted labels through ensemble voting to obtain the final predicted labels; wherein the ensemble voting method includes hard voting, soft voting, and weighted voting.

[0023] According to another aspect of the present invention, an industrial equipment fault diagnosis system based on a multi-view generation algorithm is provided, the system comprising:

[0024] The data acquisition module is configured to collect one or more operating status quantities of industrial equipment and sort the same type of operating status quantities according to time to form an original univariate time series set.

[0025] A preprocessing module configured to preprocess the original unary time series set;

[0026] The feature set generation module is configured to generate multiple views of multiple time series samples in the preprocessed original univariate time series set, and generate feature vector sets of multiple views by combining time series transformation with feature vector concatenation.

[0027] The classifier training module is configured to input the feature vector set of each view into a machine learning-based fault diagnosis classifier for training, and obtain the trained fault diagnosis classifier.

[0028] The fault prediction module is configured to generate multiple views from the time series of the preprocessed test operation state samples and obtain the corresponding feature vectors; input the feature vectors corresponding to the test operation state samples into a trained fault diagnosis classifier to obtain the fault prediction results of the test operation state samples; the specific process includes: inputting the feature vectors of K views into the corresponding trained fault diagnosis classifiers respectively, outputting K prediction labels; integrating the K prediction labels through ensemble voting to obtain the final prediction label.

[0029] Furthermore, the operating status quantities in the data acquisition module include temperature, rotational speed, and pressure; the industrial equipment includes construction machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment; the preprocessing process in the preprocessing module includes: filling time series with missing values ​​using interpolation methods; processing unequal time series into equal time series using a sliding window method; standardizing each equal time series; and labeling each standardized time series sample.

[0030] Furthermore, the specific process of generating feature vector sets for multiple views in the feature set generation module includes:

[0031] Multiple time series samples are subjected to K fast sequence transformations to obtain K transformed time series; the sequence transformations include Hilbert transform, differential transforms of various orders, wavelet transform, or Fourier transform;

[0032] Multiple time series samples and the transformed time series are transformed by random convolution kernels with kernel numbers C1 and C2 respectively, generating corresponding feature vectors. Each time series sample generates K+1 feature vectors. Specifically, each convolution kernel performs a sliding dot product operation on each time series to generate a feature map. P pooling features are extracted from each feature map, wherein the pooling features include the proportion of positive values, the proportion of negative values, the proportion of zero values, the maximum value, the minimum value, and the mean value.

[0033] The feature vectors generated from multiple time series samples and the feature vectors generated from the transformed time series are concatenated in rows to obtain feature vectors for K views. Each feature vector contains (C1+C2)P features.

[0034] Furthermore, in the feature set generation module, the time series samples s = (s1, s2, ..., s...) are transformed by random convolution kernels. n The i-th value s i The formula for the dot product operation with the convolution kernel ω is:

[0035]

[0036] Where * represents the dot product operation, d is the dilation coefficient, l represents the length of the convolution kernel ω, and ω j Let ω represent the j-th value of the convolution kernel ω, and b represent the bias coefficient.

[0037] The beneficial technical effects of this invention are:

[0038] This invention collects time-series samples using sensors installed on industrial equipment. By analyzing the collected univariate time-series sample set, it achieves fault diagnosis, determining whether the equipment has malfunctioned and the type of malfunction. This invention achieves equipment fault diagnosis through time-series classification, extracting a large number of features from the time series based on machine learning to generate feature vectors for multiple views. By combining the accuracy and diversity of multiple views with ensemble learning, the accuracy and performance of time-series classification and fault diagnosis are significantly improved.

[0039] This invention introduces various fast sequence transformations and combines them with random convolution kernel transformations to extract a large number of convolutional features from different domains. Multiple sequence transformations are applied to the time series, mapping the original sequence to different domain spaces. Then, random convolution kernel transformations are used to extract a large number of convolutional features from these different sequence spaces, thus achieving the extraction of diverse features from different domains. Feature vector concatenation is used to generate feature vectors for multiple views, ensuring the accuracy and diversity of each view. Furthermore, ensemble learning is used to synthesize the prediction results of each view, further improving classification performance. Since the accuracy of the diverse features extracted from different domains varies, with poor performance in some domains, feature vector concatenation is used to combine the feature vectors from each domain with the corresponding feature vectors from the original sequence to generate each view, thus ensuring the accuracy of each view. Ensemble learning further improves classification performance. Attached Figure Description

[0040] The present invention can be better understood by referring to the description given below in conjunction with the accompanying drawings, which together with the following detailed description are included in and form part of this specification, and are used to further illustrate preferred embodiments of the invention and explain the principles and advantages of the invention.

[0041] Figure 1 This is an overall flowchart of an industrial equipment fault diagnosis method based on a multi-view generation algorithm according to an embodiment of the present invention.

[0042] Figure 2 This is a flowchart of the data acquisition steps in an embodiment of the present invention.

[0043] Figure 3 This is a flowchart of the multi-view generation step in an embodiment of the present invention.

[0044] Figure 4 This is a flowchart illustrating the label prediction for the sample to be tested in an embodiment of the present invention. Detailed Implementation

[0045] To enable those skilled in the art to better understand the present invention, exemplary embodiments or examples of the present invention will be described below in conjunction with the accompanying drawings. Obviously, the described embodiments or examples are merely some, not all, of the embodiments or examples of the present invention. All other embodiments or examples obtained by those skilled in the art based on the embodiments or examples of the present invention without inventive effort should fall within the scope of protection of the present invention.

[0046] This invention proposes an industrial equipment fault diagnosis method and system based on a multi-view generation algorithm. The multi-view generation algorithm extracts multi-domain features and generates multi-view feature vectors by combining various fast time series transformations with convolutional features. Features from a single domain are insufficient to fully represent fault modes. By obtaining multi-domain features through multiple sequence transformations, fault features can be extracted more comprehensively. By integrating the prediction results of various views through ensemble learning, high-performance time series classification and fault diagnosis are achieved by utilizing the accuracy and diversity of each view.

[0047] This invention proposes a method for fault diagnosis of industrial equipment based on a multi-view generation algorithm, such as... Figure 1 As shown, the method includes the following steps: First, through data acquisition, the state quantities collected by a certain sensor of the industrial equipment are saved as univariate time series according to time sequence, and a large number of the same time series together form the original univariate time series set; then, through data preprocessing, the time series in the acquired univariate time series set are processed into univariate time series of equal length, equal interval, no missing values, and standardized, and each sample is labeled with the operating status of the industrial equipment, and the label includes normal and fault types; these labeled univariate time series samples form the training set; then, through multi-view generation, feature diversity is increased based on various fast time series transformations and feature vector concatenation, and the... The time series is subjected to K sequence transformations to generate corresponding transformed sequences, and convolutional features are extracted through random convolution kernel transformation to generate corresponding feature vectors. Then, the feature vectors corresponding to the original sequence are concatenated with the feature vectors corresponding to the K transformed sequences to obtain K feature vectors, i.e., K views. Next, classifiers are trained on the K views in a labeled training set to obtain K classifiers. Finally, in the test sample label prediction stage, the K view feature vectors generated from the multi-view method are input into the trained classifiers to obtain K predicted labels. Finally, through voting ensemble, the predicted label of the test sample, i.e., the operating status of the industrial equipment, is obtained. The embodiments of this invention are described in detail below.

[0048] Step 1: Data acquisition. Collect and save the original univariate time series data from the sensors of industrial equipment.

[0049] According to embodiments of the present invention, sensors measure certain operating state quantities of industrial equipment, such as temperature, rotational speed, and pressure. The industrial equipment includes, but is not limited to, construction machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment. The time series in the original univariate time series set are all collected by sensors of the same model installed at the same location on the same type of industrial equipment. The state quantities collected by a large number of sensors are sorted and saved according to time sequence to obtain the original univariate time series set.

[0050] Step 2: Data preprocessing. The collected raw univariate time series set is preprocessed to obtain a labeled training set in a unified format.

[0051] According to an embodiment of the present invention, the preprocessing process is as follows: Figure 2 As shown, it includes the following steps:

[0052] (1) Filling in missing values;

[0053] Since some time series in the original univariate time series set contain missing values, interpolation methods are used to fill in each missing value. Interpolation methods include, but are not limited to, linear interpolation, Lagrange interpolation, and spline interpolation. Through missing value filling, each time series in the original univariate time series set is now free of missing values.

[0054] (2) Divide into equal lengths;

[0055] Since the time series in the interpolated univariate time series set are of unequal length, a sliding window method is used to process them into shorter, uniform lengths to unify the sample format and facilitate processing. Through this equal-length partitioning, each time series in the interpolated univariate time series set is of equal length.

[0056] (3) Standardization process;

[0057] Since time series acquired by similar sensors all have the same unit, each time series in the univariate time series set after equal-length partitioning is standardized. For a time series of length n, s = (s1, s2, ..., s...) n The standardized formula for ) is shown in formula (1).

[0058]

[0059] Where mean(s) is the mean of time series s, and std(s) is the standard deviation of time series s.

[0060] Through standardization, each time series in the univariate time series set is processed into a standard time series with a mean of 0 and a standard deviation of 1.

[0061] (4) Labeling;

[0062] The format of the time series in the univariate time series set processed through the first three steps has been standardized. In this step, each time series sample is labeled. The label values ​​are discrete, including normal and various fault types. The time series samples are labeled according to the actual operating status of the equipment. Through labeling, the initially input raw univariate time series set is transformed into a standardized, labeled time series training set with equal lengths, equal intervals, no missing values, and no missing values. The time series training set contains m time series samples, each with a length of n.

[0063] Step 3: Multi-view generation. For each time series sample, the feature vectors of multiple views are generated by combining time series transformation with feature vector concatenation.

[0064] According to an embodiment of the present invention, the process of generating multiple views is as follows: Figure 3 As shown.

[0065] A time series sample of length n is defined as s = (s1, s2, ..., sn) n ), where s i Let be the i-th sequence value of s. First, perform K fast sequence transformations on the time series to obtain K transformed time series. The sequence transformations include, but are not limited to, Hilbert transform, differential transforms of various orders, wavelet transform, Fourier transform, etc.

[0066] Next, random convolution kernels with kernel numbers C1 and C2 are applied to both the original sequence and each transformed sequence to generate corresponding feature vectors. Each time series sample generates K+1 feature vectors. The convolution operation in the random convolution kernel transformation is shown in formula (2), where the time series sample s = (s1, s2, ..., s...). n The i-th value s i The formula for the dot product operation with the convolution kernel ω is:

[0067]

[0068] Where * represents the dot product operation, d is the dilation coefficient, l represents the length of the convolution kernel ω, and ω j Let ω represent the j-th value of the convolution kernel ω, and b represent the bias coefficient.

[0069] Each convolutional kernel performs a sliding dot product operation on the time series to generate a feature map. P pooling features are extracted from each feature map. These extracted features include, but are not limited to, the proportion of positive values, the proportion of negative values, the proportion of zero values, the maximum value, the minimum value, and the mean. Thus, each time series is transformed into K+1 feature vectors, with the feature vectors corresponding to the original sequence and the transformed sequence containing V1·P and C2·P features, respectively. The feature vectors corresponding to the original sequence are then concatenated with the feature vectors of the K transformed sequences to obtain feature vectors for K views, each containing (C1+C2)P features.

[0070] Step 4: Classifier training.

[0071] According to an embodiment of the present invention, each time series sample in the training set generates a feature vector training set of K views with a dimension of m×(C1+C2)P through the multi-view generation process in step three. A classification algorithm is trained on the training set of each view to obtain a classifier, resulting in a total of K classifiers. The classification algorithm used on each view can be the same or different. Classification algorithms include, but are not limited to, decision trees, support vector machines, logistic regression, ridge regression, k-nearest neighbors, random forests, rotation forests, Bayesian methods, and neural networks.

[0072] Step 5: Predict the label of the sample to be tested.

[0073] According to an embodiment of the present invention, the input is a time series sample of the test sample after data preprocessing, and the output is the predicted label of the sample. The prediction process is as follows: Figure 4 As shown, firstly, multi-view generation is performed to obtain feature vectors for K views. Then, the feature vectors of the K views are input into the corresponding pre-trained classifiers, which output K predicted labels. Finally, the K predicted labels are integrated through ensemble voting to obtain the final predicted label. The ensemble voting method includes, but is not limited to, hard voting, soft voting, and weighted voting.

[0074] Another embodiment of the present invention proposes an industrial equipment fault diagnosis system based on a multi-view generation algorithm, the system comprising:

[0075] The data acquisition module is configured to collect one or more operating status quantities of industrial equipment and sort the same type of operating status quantities according to time to form an original univariate time series set.

[0076] A preprocessing module configured to preprocess the original unary time series set;

[0077] The feature set generation module is configured to generate multiple views of multiple time series samples in the preprocessed original univariate time series set, and generate feature vector sets of multiple views by combining time series transformation with feature vector concatenation.

[0078] The classifier training module is configured to input the feature vector set of each view into a machine learning-based fault diagnosis classifier to obtain a trained fault diagnosis classifier.

[0079] The fault prediction module is configured to generate multiple views from the time series of the preprocessed test operation state samples and obtain the corresponding feature vectors; input the feature vectors corresponding to the test operation state samples into a trained fault diagnosis classifier to obtain the fault prediction results of the test operation state samples; the specific process includes: inputting the feature vectors of K views into the corresponding trained fault diagnosis classifiers respectively, outputting K prediction labels; integrating the K prediction labels through ensemble voting to obtain the final prediction label.

[0080] Furthermore, the operating status quantities in the data acquisition module include temperature, rotational speed, and pressure; the industrial equipment includes construction machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment; the preprocessing process in the preprocessing module includes: filling time series with missing values ​​using interpolation methods; processing unequal time series into equal time series using a sliding window method; standardizing each equal time series; and labeling each standardized time series sample.

[0081] Furthermore, the specific process of generating feature vector sets for multiple views in the feature set generation module includes:

[0082] Multiple time series samples are subjected to K fast sequence transformations to obtain K transformed time series; the sequence transformations include Hilbert transform, differential transforms of various orders, wavelet transform, or Fourier transform;

[0083] Multiple time series samples and the transformed time series are transformed by random convolutional kernels with kernel numbers C1 and C2, respectively, to generate corresponding feature vectors. Each time series sample generates K+1 feature vectors. Specifically, this includes: each convolutional kernel performing a sliding dot product operation on each time series to generate a feature map; and extracting P pooling features from each feature map, wherein the pooling features include the proportion of positive values, the proportion of negative values, the proportion of zero values, the maximum value, the minimum value, and the mean value.

[0084] The feature vectors generated from multiple time series samples and the feature vectors generated from the transformed time series are concatenated to obtain feature vectors for K views. Each feature vector contains (C1+C2)P features.

[0085] Furthermore, in the feature set generation module, the time series samples s = (s1, s2, ..., s...) are transformed by random convolution kernels. n The i-th value s iThe formula for the dot product operation with the convolution kernel ω is:

[0086]

[0087] Where * represents the dot product operation, d is the dilation coefficient, l represents the length of the convolution kernel ω, and ω j Let ω represent the j-th value of the convolution kernel ω, and b represent the bias coefficient.

[0088] The functionality of the industrial equipment fault diagnosis system based on a multi-view generation algorithm in this embodiment of the invention can be described by the aforementioned industrial equipment fault diagnosis method based on a multi-view generation algorithm. Therefore, for the parts not detailed in the system embodiment, please refer to the above method embodiment, and they will not be repeated here.

[0089] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. The disclosure of the invention is illustrative and not restrictive, and the scope of the invention is defined by the appended claims.

Claims

1. A method for industrial equipment fault diagnosis based on a multi-view generation algorithm, characterized in that, The method comprises the following steps: Step one, collecting one or more operating state quantities of industrial equipment, and sorting the same operating state quantities according to time to form an original one-dimensional time series set; Step two, preprocessing the original one-dimensional time series set; Step three, multi-view generation is performed on the plurality of time series samples in the pre-processed original unary time series set, and a plurality of view feature vector sets are generated through the combination of time series transformation and feature vector splicing; including: performing each time series sample A fast sequence transformation is performed to obtain transformed time series; the sequence transformation includes Hilbert transformation, differential transformation of each order, wavelet transformation or Fourier transformation; The random convolution kernel transformation with a kernel number of is performed on each time series sample and the transformed time series respectively to generate corresponding feature vectors, and feature vectors are generated for each time series sample; specifically comprising: each convolution kernel performs a sliding dot product operation on each time series to generate a feature map, and pooling features are extracted from each feature map, wherein the pooling features include the proportion of positive values, the proportion of negative values, the proportion of zero values, the maximum value, the minimum value, and the average value; The feature vectors corresponding to each time series sample and the feature vectors corresponding to the transformed time series are spliced in rows to obtain a feature vector of each view, each feature vector containing features. Step four, inputting the feature vector set of each view into a machine learning-based fault diagnosis classifier for training to obtain a trained fault diagnosis classifier; Step five, generating a multi-view for the preprocessed time series of the to-be-tested operating state sample to obtain the corresponding feature vector; Step six, inputting the feature vector corresponding to the to-be-tested operating state sample into the trained fault diagnosis classifier to obtain the fault prediction result of the to-be-tested operating state sample.

2. The industrial equipment fault diagnosis method based on multi-view generation algorithm according to claim 1, characterized in that, The operating state quantity in step one includes temperature, speed, and pressure; and the industrial equipment includes engineering machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment. 3.The industrial equipment fault diagnosis method based on multi-view generation algorithm of claim 1, wherein, The preprocessing in step two includes: filling the time series with missing values by using an interpolation method; processing unequal-length time series into equal-length time series by using a sliding window method; standardizing each equal-length time series; and marking a label for each time series sample after standardization.

4. The industrial equipment fault diagnosis method based on a multi-view generation algorithm according to claim 1, characterized in that, Time series samples in random kernel transformation in step three The first value and the dot product operation formula of the kernel is: ； wherein, is a dot product operation, is a dilation coefficient, denotes a length of the convolution kernel denotes a length of the convolution kernel denotes a value of the convolution kernel denotes a value of the convolution kernel denotes a value of the convolution kernel denotes a bias coefficient.

5. The industrial equipment fault diagnosis method based on multi-view generation algorithm according to claim 1, characterized in that, The specific process in step six includes: inputting the feature vectors of the views into corresponding trained fault diagnosis classifiers respectively, outputting predicted labels; integrating the predicted labels by ensemble voting to obtain a final predicted label; wherein the method of ensemble voting includes hard voting, soft voting, and weighted voting.

6. An industrial equipment failure diagnosis system based on a multi-view generation algorithm, characterized by, The method comprises: a data acquisition module configured to collect one or more operating state quantities of industrial equipment, and sort the same operating state quantities according to time to form an original one-dimensional time series set; a preprocessing module configured to preprocess the original one-dimensional time series set; a feature set generation module configured to generate a multi-view for a plurality of time series samples in the preprocessed original one-dimensional time series set, and combine time series transformation and feature vector splicing to generate a feature vector set of a plurality of views; The specific process comprises: obtaining a transformed time sequence by performing a fast sequence transformation on each time sequence sample The sequence transformation comprises a Hilbert transformation, a differential transformation of each order, a wavelet transformation or a Fourier transformation. The random convolution kernel transformation is performed on each time sequence sample and the transformed time sequence respectively, and a corresponding feature vector is generated, and each time sequence sample generates The specific process comprises: each convolution kernel performs a sliding dot product operation on each time sequence to generate a feature map, and a plurality of pooling features are extracted from each feature map, wherein the pooling features comprise a proportion of positive values, a proportion of negative values, a proportion of zero values, a maximum value, a minimum value and an average value. The feature vectors of the plurality of views are obtained by splicing the feature vectors corresponding to each time sequence sample and the feature vectors corresponding to the transformed time sequence in rows, and each feature vector contains a plurality of features. a classifier training module configured to input the feature vector set of each view into a machine learning-based fault diagnosis classifier for training to obtain a trained fault diagnosis classifier; The fault prediction module is configured to perform multi-view generation on the time series of the preprocessed to-be-tested running state sample to obtain a corresponding feature vector; input the feature vector corresponding to the to-be-tested running state sample into the trained fault diagnosis classifier to obtain a fault prediction result of the to-be-tested running state sample; and the specific process comprises: inputting the feature vectors of the plurality of views into the corresponding trained fault diagnosis classifiers respectively, and outputting prediction labels; and integrating the prediction labels through ensemble voting to obtain a final prediction label.

7. The industrial equipment fault diagnosis system based on a multi-view generation algorithm according to claim 6, characterized in that, The operating state quantity in the data acquisition module includes temperature, speed, and pressure; and the industrial equipment includes engineering machinery, vehicles, machine tools, environmental protection machinery, electrical equipment, and electronic equipment. The preprocessing process in the preprocessing module includes: filling the time series with missing values by using an interpolation method; processing unequal-length time series into equal-length time series by using a sliding window method; standardizing each equal-length time series; and marking a label for each time series sample after standardization.

8. The industrial equipment fault diagnosis system based on a multi-view generation algorithm according to claim 6, characterized in that, The time sequence sample in random convolution kernel transformation in the feature set generation module The first The value The dot product operation formula of the convolution kernel is: ； in, For dot product operation, The coefficient of thermal expansion is 1 / 3. Represents the convolution kernel Length, Represents the convolution kernel The One value, This represents the deviation coefficient.

Images