A method, system, device, and medium for assessing equipment reliability.

Abstract

This invention relates to the field of equipment management technology and discloses a method, system, device, and medium for equipment reliability assessment. The method employs a deep autoencoder and a self-attention mechanism encoder to extract features from a real-time time-series dataset reflecting the state of the equipment under test, yielding potential key features and temporal features. A decoder reconstructs the first fused feature obtained from the fusion of the potential key features and temporal features, resulting in a reconstructed sample. The reconstruction error of the reconstructed sample is calculated, and the reconstruction error is fused with the first fused feature to obtain a second fused feature. This second fused feature is input into a Gaussian mixture model, and the resulting real-time negative log-likelihood value is compared with a preset threshold to obtain a deviation value. Based on the deviation value, a preset reliability assessment algorithm is used for evaluation to obtain the reliability assessment result of the equipment under test. The method of this application effectively improves the accuracy of equipment reliability assessment.

Description

Technical Field

[0001] This invention relates to the field of equipment management technology, and in particular to a method, system, equipment and medium for equipment reliability assessment. Background Technology

[0002] In the industrial sector, timely and accurate identification of abnormal equipment operation and scientific assessment of equipment reliability are core aspects of ensuring safe and stable equipment operation and improving maintenance efficiency. With the development of industrial automation and intelligence, abnormal equipment operation can be detected by analyzing large amounts of real-time collected equipment operation time-series data, and reliability assessments can be performed based on the detection results.

[0003] In existing technologies, deep autoencoders are commonly used to model and analyze time-series data of equipment operation to achieve real-time monitoring and anomaly detection, thereby assessing equipment reliability. While deep autoencoders offer advantages in anomaly reconstruction and feature compression, this model also has significant drawbacks. It often ignores key temporal correlations in time-series data, leading to response lag and prediction distortion during anomaly detection. This makes it difficult to quickly and accurately identify early equipment anomalies, resulting in low accuracy in equipment reliability assessments.

[0004] Therefore, improving the accuracy of equipment reliability assessment has become a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This invention provides a method, system, device, and medium for equipment reliability assessment, in order to solve the technical problem of how to improve the accuracy of equipment reliability assessment, realize dynamic assessment and prediction of equipment reliability, and improve the accuracy of equipment reliability assessment.

[0006] In a first aspect, the present invention provides a method for evaluating equipment reliability, the method comprising:

[0007] An initial real-time time series dataset that reflects the status of the device under test is collected, and the initial real-time time series dataset is preprocessed and standardized to obtain the target real-time time series dataset.

[0008] A trained deep autoencoder is used to extract potential key features from the target real-time time series dataset to obtain potential key features. A trained self-attention mechanism encoder is used to extract temporal features from the target real-time time series dataset to obtain temporal features. The potential key features and the temporal features are fused to obtain a first fused feature.

[0009] The first fusion feature is reconstructed using the trained decoder to obtain a reconstructed sample, and the reconstruction error of the reconstructed sample is calculated. The reconstruction error is then fused with the first fusion feature to obtain a second fusion feature.

[0010] The second fusion feature is input into the trained Gaussian mixture model to obtain a real-time negative log-likelihood value. The real-time negative log-likelihood value is then compared with a preset threshold to obtain a deviation value. Based on the deviation value, a preset reliability assessment algorithm is used to evaluate the reliability of the device under test, thereby obtaining the reliability assessment result of the device under test.

[0011] Preferably, the step of using a trained self-attention mechanism encoder to extract temporal features from the target real-time time series dataset to obtain temporal features includes:

[0012] Location codes are added to the real-time time series data of the target real-time time series dataset to obtain the location information of the real-time time series data;

[0013] A trained self-attention mechanism encoder is used to extract temporal features from the target real-time time series dataset and the location information to obtain temporal features.

[0014] Preferably, calculating the reconstruction error of the reconstructed sample includes:

[0015] The cosine similarity algorithm is used to calculate the cosine distance between the target real-time time series dataset and the reconstructed sample to obtain the cosine similarity error;

[0016] The Euclidean distance algorithm is used to calculate the Euclidean distance between the target real-time time series dataset and the reconstructed sample to obtain the Euclidean distance error;

[0017] The cosine similarity error and the Euclidean distance error are fused to obtain the reconstruction error.

[0018] Preferably, obtaining the preset threshold includes:

[0019] A comprehensive loss function is set for the constructed deep autoencoder, the self-attention mechanism encoder, the decoder, and the Gaussian mixture model;

[0020] The first historical time series dataset of the device to be detected is collected. Based on the comprehensive loss function and the first historical time series dataset, the deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model are trained end-to-end to obtain the trained deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model.

[0021] A second historical time series dataset of the normal state of the device to be detected is collected. Based on the second historical time series dataset, the deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model are used to perform corresponding processing to obtain a set of negative log-likelihood values ​​of the normal state.

[0022] The set of negative log-likelihood values ​​for the normal state is analyzed, and a preset threshold is obtained based on the analysis results.

[0023] Preferably, the comprehensive loss function is expressed as:

[0024]

[0025]

[0026] in, Represents the loss function. This indicates the calculation of the mean square error. This represents the initial input dataset. This represents the reconstructed samples obtained based on the initial input dataset. This represents the input to the Gaussian mixture model. This represents the output of the Gaussian mixture model. This represents the output weights of the Gaussian mixture model. Represents a regular term, Indicates the weight of the regularization term. The first Gaussian mixture model obtained by fitting the model represents the... Gaussian sub-distribution, The input of the Gaussian mixture model represents the first... One characteristic, This represents the total number of Gaussian sub-distributions obtained by fitting the Gaussian mixture model. This represents the feature dimension of the input to the Gaussian mixture model. Indicates the first The first Gaussian sub-distribution's covariance matrix The diagonal elements corresponding to each feature.

[0027] Preferably, the reliability assessment algorithm consists of a risk proportion function and a reliability function;

[0028] The risk proportion function is expressed as follows:

[0029]

[0030] in, Indicates time, Indicates the deviation value. express Time deviation value The risk rate below Represents the benchmark risk function. Represents the fitting coefficient. Represented by natural constant An exponential function with base 0;

[0031] The reliability function is expressed as follows:

[0032]

[0033] in, express Time deviation value The reliability of the equipment is as follows.

[0034] Preferably, the preprocessing and standardization of the initial real-time time series dataset to obtain the target real-time time series dataset includes:

[0035] The initial real-time time series dataset is subjected to correlation analysis using the maximum information coefficient, and the real-time time series dataset is filtered based on the correlation analysis results to obtain the filtered initial real-time time series dataset.

[0036] The initial real-time time series dataset after filtering is normalized to obtain the target real-time time series dataset.

[0037] Secondly, the present invention also provides a device reliability assessment system to implement the device reliability assessment method described above. The system includes: a data preprocessing unit, a feature extraction unit, a reconstruction error fusion unit, and a reliability assessment unit.

[0038] The data preprocessing unit is used to collect an initial real-time time series dataset that reflects the status of the device to be detected, and to preprocess and standardize the initial real-time time series dataset to obtain a target real-time time series dataset.

[0039] The feature extraction unit is used to extract potential key features from the target real-time time series dataset using a trained deep autoencoder to obtain potential key features, to extract temporal features from the target real-time time series dataset using a trained self-attention mechanism encoder to obtain temporal features, and to fuse the potential key features and the temporal features to obtain a first fused feature.

[0040] The reconstruction error fusion unit is used to reconstruct the first fusion feature using the trained decoder to obtain a reconstructed sample, calculate the reconstruction error of the reconstructed sample, and fuse the reconstruction error with the first fusion feature to obtain a second fusion feature.

[0041] The reliability assessment unit is used to input the second fused feature into the trained Gaussian mixture model to obtain a real-time negative log-likelihood value, compare the real-time negative log-likelihood value with a preset threshold to obtain a deviation value, and evaluate the reliability of the device under test using a preset reliability assessment algorithm based on the deviation value to obtain the reliability assessment result of the device under test. The reliability assessment algorithm is set to consist of a risk proportion function and a reliability function.

[0042] Thirdly, the present invention also provides a computer device, the computer device including a memory, a processor and a transceiver, which are connected to each other via a bus; the memory is used to store a set of computer program instructions and data, and to transmit the stored data to the processor, the processor executing the computer program instructions stored in the memory to perform the device reliability assessment method described above.

[0043] Fourthly, the present invention also provides a computer-readable storage medium storing a computer program that, when executed, implements the device reliability assessment method described above.

[0044] This application provides a method, system, device, and medium for evaluating equipment reliability. Compared with the prior art, the beneficial effects of the embodiments of this application are as follows:

[0045] The equipment reliability assessment method disclosed in this application combines a deep autoencoder with a self-attention mechanism encoder. This not only acquires high-order coupling features between time-series data but also obtains dependency and trend features, thereby introducing dynamic temporal features and improving the accuracy of subsequent equipment reliability assessments. It also introduces reconstruction errors to broaden the distinction between normal and abnormal samples, enhancing the ability of subsequent Gaussian mixture models to differentiate between them. Furthermore, it obtains the real-time negative log-likelihood value of the second fusion feature through the Gaussian mixture model, achieving higher accuracy and stronger explanatory power for anomaly detection. Finally, it performs equipment reliability assessment based on the deviation between the real-time negative log-likelihood value and a preset threshold, realizing dynamic assessment and prediction of equipment reliability and improving the accuracy of equipment reliability assessments. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the steps of a device reliability assessment method provided in a preferred embodiment of the present invention;

[0047] Figure 2 This is an abnormal detection result of a superheater leakage event of a certain unit, provided by a preferred embodiment of the present invention, using the equipment reliability assessment method disclosed in this application;

[0048] Figure 3This is a preferred embodiment of the present invention, which provides a reliability assessment result of a superheater obtained using the equipment reliability assessment method disclosed in this application;

[0049] Figure 4 This is a schematic diagram of the structure of a device reliability assessment system provided in a preferred embodiment of the present invention;

[0050] Figure 5 This is an internal structural diagram of the computer device in an embodiment of the present invention;

[0051] Figure label:

[0052] 1-Data preprocessing unit, 2-Feature extraction unit, 3-Reconstruction error fusion unit, 4-Reliability assessment unit. Detailed Implementation

[0053] The embodiments of the present invention are described in detail below with reference to the accompanying drawings. The embodiments are provided for illustrative purposes only and should not be construed as limiting the invention. The accompanying drawings are for reference and illustration only and do not constitute a limitation on the scope of protection of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of this invention. In the description of this invention, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this invention, unless otherwise stated, "a plurality of" means two or more.

[0054] In the description of this invention, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to communication within two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art will understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0055] In the description of this invention, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0056] Please see Figure 1 In an embodiment of the present invention, a device reliability assessment method is provided, the method comprising:

[0057] S1. Collect an initial real-time time series dataset that reflects the state of the equipment under test, and preprocess and standardize the initial real-time time series dataset to obtain a target real-time time series dataset; In a preferred embodiment of this application, during the operation of the equipment under test, an initial real-time time series dataset of parameters related to the equipment state is collected. The parameters reflecting the equipment state include mechanical parameters, electrical parameters, process and environmental parameters, and operating status and control parameters. Among them, mechanical parameters include at least: vibration amplitude, vibration frequency and phase of vibration signal, bearing temperature, motor temperature, gearbox temperature, speed and torque, oil condition data, wear debris data, etc.; electrical data include at least: The initial real-time time series dataset includes data on current, voltage, power, energy efficiency, insulation performance, insulation resistance, and partial discharge. Process and environmental parameters include at least pressure, flow rate, liquid level, process temperature, humidity, dust concentration, and noise. Operating status and control parameters include at least start / stop status, operating mode, cumulative running time, setpoint, feedback value, alarm and fault codes. Other specialized parameters are customized based on equipment type or industry characteristics, such as compressor suction / discharge pressure, volumetric efficiency, and surge warning parameters; boiler steam pressure, water level, and flue gas composition; CNC machine tool spindle accuracy, tool wear, and machining dimensional deviations; and robot joint angles, torque feedback, and path errors. Therefore, the types of data included in the initial real-time time series dataset should be determined specifically according to the type of equipment, and no specific limitations are made in this application.

[0058] For the initial real-time time series dataset, correlation analysis was performed using the maximum information coefficient. Redundant data was removed based on the correlation analysis results. First, each data point in the initial real-time time series dataset was binned, and its value range was divided into... The data is divided into grids. Data binning transforms continuous numerical data into discrete intervals, or bins. For each grid division, the mutual information of the discretized data is calculated, and the maximum mutual information value is found among all possible grid divisions to obtain the maximum information coefficient.

[0059] The formula for calculating the maximum information coefficient is:

[0060]

[0061]

[0062] in, and The constraint representing the total number of grid divisions. and This represents any two classes of data in the initial real-time time series dataset. Represents a grid The maximum number of information items, Represents a grid. Denotes the joint probability density function. Representing data The marginal probability density function, Representing data The marginal probability density function, This represents the maximum mutual information coefficient after normalization, and B represents the constraint value, which is usually taken as 0.6 times the total number of data points.

[0063] Based on the calculated maximum mutual information coefficient, high-value features are screened from the initial real-time time series dataset to remove redundant or irrelevant data, thereby improving the data's noise robustness and anti-interference ability, and providing a high-value dataset for subsequent data analysis and modeling.

[0064] Furthermore, the initial real-time time series data after screening is normalized to ensure that the initial real-time time series data have the same dimensions, thus obtaining the target real-time time series data.

[0065] S2. A trained deep autoencoder is used to extract latent features from the target real-time time series dataset to obtain latent features. A trained self-attention mechanism encoder is used to extract temporal features from the target real-time time series dataset to obtain temporal features. The latent features and the temporal features are fused to obtain a first fused feature. In a preferred embodiment of this application, in order to overcome the problem that the deep autoencoder ignores the key temporal correlations in the time series data, resulting in response lag and prediction distortion during sequence anomaly detection, a self-attention mechanism encoder is introduced, and features are extracted from the target real-time time series dataset in combination with the deep autoencoder.

[0066] A deep autoencoder is structured as a multi-layer fully connected neural network, learning complex representations of data using multiple hidden layers. In a preferred embodiment of this application, a denoising autoencoder (DAE) is selected. The DAE receives noisy input data and encodes it into a representation in a low-dimensional space. In this process, the DAE attempts to learn a feature representation that allows the subsequent decoder to reconstruct an output close to the original clean data, even if the input data is noisy.

[0067] The target real-time time series dataset is input into the trained deep autoencoder. The deep autoencoder extracts potential key features from the target real-time time series dataset, which mainly include high-order coupling features between real-time time series data in the target real-time time series dataset.

[0068] In a preferred embodiment of this application, the self-attention mechanism encoder is a Transformer encoder. The Transformer encoder consists of multiple identical coding blocks stacked together, typically six layers. Each coding block contains a multi-head self-attention mechanism and a feedforward neural network. The multi-head self-attention mechanism first linearly transforms the input target real-time time series dataset into a query vector, a key vector, and a value vector. Each "head" of the multi-head self-attention mechanism independently calculates the dot product between the query vector and the key vector, normalizes it using a learnable scaling factor, and obtains an attention score. This attention score represents the correlation between different positions in the input target real-time time series dataset. Each "head" performs a weighted summation of the value vectors based on its attention score to obtain its output. Finally, the outputs of all "heads" are concatenated and merged using a linear transformation matrix to obtain a more comprehensive semantic representation. The feedforward neural network is a simple two-layer fully connected network with a non-linear activation function sandwiched in between. Its function is to further process the output of the self-attention mechanism, introduce non-linear transformations, and enhance the model's expressive power.

[0069] Before using the trained self-attention mechanism encoder to extract temporal features from the target real-time time series dataset, position encoding needs to be added to the real-time time series data of the target real-time time series dataset to obtain the position information of each time step of the time series data, so that the self-attention mechanism encoder can perceive the time order when processing the real-time time series data.

[0070] The position encoding process is represented as follows:

[0071]

[0072] in, Indicates the index of the current time step. Indicates the feature dimension index. Represents the feature dimension of the encoding. and They represent respectively to Feature Dimensions When Encoding Location at Time Steps and The information above.

[0073] Furthermore, a multi-head attention mechanism based on a self-attention encoder constructed from a Transformer encoder is used to process the real-time time series dataset of the target with added location information to obtain the correlation between different time steps. The calculation process of the multi-head attention mechanism is as follows:

[0074]

[0075]

[0076]

[0077] in, , , These represent the query vector, key vector, and value vector, respectively. , , These represent the query vector weight matrix, key vector weight matrix, and key vector weight matrix, respectively, for the multi-head attention mechanism. This represents the activation function. This represents the dimension of the key vector. Indicates the first The output of each attention head, Indicates the first The output of each attention head, This indicates the number of heads in the multi-head attention mechanism. The weight matrix representing multi-head attention. This indicates multi-head output.

[0078] After processing by the multi-head attention mechanism, the output of the multi-head attention mechanism is processed by a feedforward neural network for residual connection and normalization to obtain the temporal features. The calculation process is as follows:

[0079]

[0080]

[0081] in, For normalization processing, For the normalized result, This is for fully connected layer processing. It represents the temporal characteristics.

[0082] Furthermore, the potential key features and time series features are fused to obtain the first fused feature. The first fused feature includes not only the high-order coupling features between real-time time series data in the target real-time time series dataset, but also the dependency relationship features and change trend features between real-time time series data in the target real-time time series dataset, so as to introduce dynamic time series features and improve the accuracy of subsequent equipment reliability assessment.

[0083] S3. The first fusion feature is reconstructed using the trained decoder to obtain a reconstructed sample, and the reconstruction error of the reconstructed sample is calculated. The reconstruction error and the first fusion feature are then fused to obtain a second fusion feature. In a preferred embodiment of this application, the first fusion feature is reconstructed using a decoder composed of a multi-layer fully connected neural network. The first fusion feature is input into the trained decoder for sample reconstruction to obtain a reconstructed sample. Further, the reconstruction error of the reconstructed sample is calculated. In this application, the cosine similarity algorithm and the Euclidean distance algorithm are used to calculate the reconstruction error respectively. The cosine similarity algorithm is used to calculate the cosine distance between the target real-time time series dataset and the reconstructed sample to obtain the cosine similarity error. The calculation formula is as follows:

[0084]

[0085] in, This represents the cosine similarity error. This represents the initial input dataset for the deep autoencoder, i.e., the target real-time time series dataset. Indicates the reconstructed sample. It represents the 2-norm.

[0086] The Euclidean distance algorithm is used to calculate the Euclidean distance between the target real-time time series dataset and the reconstructed samples, and the Euclidean distance error is obtained. The calculation formula is as follows:

[0087]

[0088] in, This indicates the Euclidean distance error.

[0089] Finally, the cosine similarity error and the Euclidean distance error are fused to obtain the reconstruction error. In this application, the commonly used additive fusion method is selected, which is simple and intuitive.

[0090] Furthermore, the reconstruction error and the first fusion feature are further fused to obtain the second fusion feature, which amplifies the distinction between normal and abnormal samples. Abnormal samples have weak reconstruction capabilities and large cosine similarity and Euclidean distance errors. Therefore, incorporating the reconstruction error as part of the final feature used for equipment reliability assessment can amplify the distinction between normal and abnormal samples, improve the ability of the subsequent Gaussian mixture model to differentiate between normal and abnormal samples, and enhance the accuracy of equipment reliability assessment.

[0091] S4. Input the second fusion feature into the trained Gaussian mixture model to obtain the real-time negative log-likelihood value. Compare the real-time negative log-likelihood value with a preset threshold to obtain the deviation value. Based on the deviation value, use a preset reliability assessment algorithm to evaluate and obtain the reliability assessment result of the device under test. Existing technologies generally rely on outlier judgment results of device operation time series data to judge the abnormal state of the device. However, relying solely on outlier judgment results cannot provide comprehensive and systematic support for device reliability assessment. In the preferred embodiment of this application, a Gaussian mixture model is used to process the second fusion feature. A Gaussian mixture model is a probability-based statistical model used to describe a mixed data distribution composed of multiple Gaussian distributions. Its core idea is to assume that the data is generated by a mixture of several implicit Gaussian sub-distributions, and each Gaussian sub-distribution corresponds to a latent structure in the data. By estimating the mean, covariance, and weights of the parameters of these Gaussian sub-distributions, tasks such as modeling, cluster analysis, and density estimation of complex data distributions can be achieved. Assuming the constructed Gaussian mixture model contains K Gaussian sub-distributions, each corresponding to a Gaussian distribution, the output of the Gaussian mixture model is the negative log-likelihood value obtained after processing the input data. It can more accurately describe the global distribution structure of normal data, avoid misjudging modal differences within the normal range as anomalies, achieve continuous quantification of the degree of anomalies, and support hierarchical early warning.

[0092] In the Gaussian mixture model processing, an evaluation network consisting of multiple fully connected layers is first used to calculate the input, obtaining the corresponding probability distribution. The calculation formula is as follows:

[0093]

[0094] in, To evaluate the probability distribution of the network output, This represents the input to the Gaussian mixture model. This represents a nonlinear function that transforms any real-valued vector into a probability distribution.

[0095] Furthermore, the parameters of the Gaussian mixture model are estimated based on the probability distribution, and the calculation formula is as follows:

[0096]

[0097]

[0098]

[0099] in, The first Gaussian mixture model obtained by fitting the model represents the... Gaussian sub-distribution, In the Gaussian mixture model, the first... The mixture probability of a Gaussian sub-distribution This represents the total number of samples input to the Gaussian mixture model. Indicates the first One sample, Indicates the first The sample belongs to the first The probability of a Gaussian sub-distribution. In the Gaussian mixture model, the first... The mean of a Gaussian sub-distribution, Indicates the first Features of each sample In the Gaussian mixture model, the first... The covariance of a Gaussian sub-distribution.

[0100] Furthermore, the Gaussian mixture model input is processed using the parameter-estimated Gaussian mixture model to calculate the negative log-likelihood value. The calculation formula is as follows:

[0101]

[0102] in, This represents the output of the Gaussian mixture model, specifically the negative log-likelihood value. This represents the feature dimension of the Gaussian mixture model input.

[0103] In a preferred embodiment of this application, the preset threshold is obtained by processing the historical time-series dataset of the normal state of the device to be detected. Specifically, a comprehensive loss function is set for the constructed deep autoencoder, self-attention mechanism encoder, decoder, and Gaussian mixture model. The comprehensive loss function set for the constructed deep autoencoder, self-attention mechanism encoder, decoder, and Gaussian mixture model is expressed as follows:

[0104]

[0105]

[0106] in, Represents the loss function. This indicates the calculation of the mean square error. This represents the initial input dataset. This represents the reconstructed samples obtained based on the initial input dataset. This represents the input to the Gaussian mixture model. This represents the output of the Gaussian mixture model. This represents the output weights of the Gaussian mixture model. Represents a regular term, Indicates the weight of the regularization term. The input of the Gaussian mixture model represents the first... One characteristic, This represents the total number of Gaussian sub-distributions obtained by fitting the Gaussian mixture model. This represents the feature dimension of the input to the Gaussian mixture model. Indicates the first The first Gaussian sub-distribution's covariance matrix The diagonal elements corresponding to each feature.

[0107] Furthermore, the first historical time series data of the device under test is collected. Based on the comprehensive loss function and the first historical time series data, the deep autoencoder, self-attention mechanism encoder, decoder and Gaussian mixture model are trained end-to-end to obtain the trained deep autoencoder, self-attention mechanism encoder, decoder and Gaussian mixture model.

[0108] Furthermore, a second historical time-series dataset of the device under test in its normal state is collected. This dataset is then processed and fused using a trained deep autoencoder, self-attention encoder, decoder, and Gaussian mixture model to obtain a set of negative log-likelihood values ​​for the normal state corresponding to the second historical time-series dataset. This set of negative log-likelihood values ​​for the normal state is analyzed, and a preset threshold is obtained based on the analysis results. This application employs mean analysis to calculate the mean of the set of negative log-likelihood values ​​for the normal state, and uses this mean as the preset threshold.

[0109] Furthermore, the second fusion feature obtained from the initial real-time time series dataset is input into the trained Gaussian mixture model to obtain the real-time negative log-likelihood value. The real-time negative log-likelihood value is then compared with a preset threshold to obtain the deviation value. The deviation value is used as a covariate, and a reliability assessment algorithm is used for prediction to obtain the reliability assessment result of the device under test. The reliability assessment algorithm is set to consist of a risk proportion function and a reliability function, wherein the risk proportion function is expressed as follows:

[0110]

[0111] in, Indicates time, Indicates the deviation value. express Time deviation value The risk rate below Represents the benchmark risk function. Represents the fitting coefficient. Represented by natural constant An exponential function with base 0;

[0112] The reliability function is expressed as follows:

[0113]

[0114] in, express Time deviation value The reliability of the equipment is as follows.

[0115] By using a Gaussian mixture model to obtain the real-time negative log-likelihood value of the second fusion feature, higher accuracy and stronger interpretability of equipment status anomaly detection are achieved. Furthermore, based on the deviation between the real-time negative log-likelihood value and a preset threshold, equipment reliability is assessed, enabling dynamic evaluation and prediction of equipment reliability. This improves the safety and economy of industrial systems, prevents major accidents, and helps maintenance personnel respond promptly to emergencies.

[0116] In one embodiment of this application, as Figure 2 The image shows the abnormal detection results of a superheater leak incident in a certain unit using the equipment reliability assessment method disclosed in this application. In reality, staff heard an abnormal noise at 6:49 AM on May 7th, but observed that all unit sensor data were normal. It was later confirmed that the leak was in a pipeline. Figure 2 As can be seen, the equipment reliability assessment method disclosed in this application was used for testing, and it was found that the real-time negative log-likelihood value exceeded the preset threshold starting at 1:00 AM on May 7th, approximately 6 hours earlier than the time the operators discovered it, proving that the present invention can detect equipment anomalies in advance. Figure 3 The figure shows the reliability assessment results of the superheater obtained using the equipment reliability assessment method disclosed in this application. Reliability is used to measure equipment reliability. The equipment reliability shows a significant decrease around point 1, and a rapid decrease around point 6:20, which is consistent with actual results. Therefore, this invention can effectively assess and monitor equipment reliability.

[0117] In a preferred embodiment of the present invention, an initial real-time time series dataset reflecting the state of the device under test is collected, and the initial real-time time series dataset is preprocessed and standardized to obtain a target real-time time series dataset. A trained deep autoencoder is used to extract potential key features from the target real-time time series dataset to obtain potential key features. A trained self-attention mechanism encoder is used to extract temporal features from the target real-time time series dataset to obtain temporal features. The potential key features and temporal features are fused to obtain a first fused feature. A trained decoder is used to reconstruct the first fused feature to obtain reconstructed samples, and the reconstruction error of the reconstructed samples is calculated. The reconstruction error and the first fused feature are fused to obtain a second fused feature. The second fused feature is input into a trained Gaussian mixture model to obtain a real-time negative log-likelihood value. The real-time negative log-likelihood value is compared with a preset threshold to obtain a deviation value. Based on the deviation value, a preset reliability assessment algorithm is used for evaluation to obtain the reliability assessment result of the device under test. The equipment reliability assessment method disclosed in this application combines a deep autoencoder with a self-attention mechanism encoder. This not only acquires high-order coupling features between time-series data but also obtains dependency and trend features, thereby introducing dynamic temporal features and improving the accuracy of subsequent equipment reliability assessments. It also introduces reconstruction errors to broaden the distinction between normal and abnormal samples, enhancing the ability of subsequent Gaussian mixture models to differentiate between them. Furthermore, it obtains the real-time negative log-likelihood value of the second fusion feature through the Gaussian mixture model, achieving higher accuracy and stronger explanatory power for anomaly detection. Finally, it performs equipment reliability assessment based on the deviation between the real-time negative log-likelihood value and a preset threshold, realizing dynamic assessment and prediction of equipment reliability and improving the accuracy of equipment reliability assessments.

[0118] Accordingly, such as Figure 4 As shown, based on a device reliability assessment method, this embodiment of the invention also provides a device reliability assessment system to implement the device reliability assessment method disclosed in this embodiment of the invention. The system includes: a data preprocessing unit 1, a feature extraction unit 2, a reconstruction error fusion unit 3, and a reliability assessment unit 4.

[0119] The data preprocessing unit 1 is used to collect an initial real-time time series dataset that reflects the status of the device to be detected, and to preprocess and standardize the initial real-time time series dataset to obtain a target real-time time series dataset.

[0120] The feature extraction unit 2 is used to extract potential key features from the target real-time time series dataset using a trained deep autoencoder to obtain potential key features, extract temporal features from the target real-time time series dataset using a trained self-attention mechanism encoder to obtain temporal features, and fuse the potential key features and the temporal features to obtain a first fused feature.

[0121] The reconstruction error fusion unit 3 is used to reconstruct the first fusion feature using the trained decoder to obtain a reconstructed sample, calculate the reconstruction error of the reconstructed sample, and fuse the reconstruction error with the first fusion feature to obtain a second fusion feature.

[0122] The reliability assessment unit 4 is used to input the second fused feature into the trained Gaussian mixture model to obtain a real-time negative log-likelihood value, compare the real-time negative log-likelihood value with a preset threshold to obtain a deviation value, and evaluate the reliability of the device under test using a preset reliability assessment algorithm based on the deviation value to obtain the reliability assessment result of the device under test.

[0123] For specific limitations regarding a device reliability assessment system, please refer to the above-described limitations regarding a device reliability assessment method, which will not be repeated here. Those skilled in the art will recognize that the various modules and steps described in conjunction with the embodiments disclosed in this invention can be implemented in hardware, software, or a combination of both. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention.

[0124] like Figure 5 As shown, an embodiment of the present invention provides a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps as described in the device reliability assessment embodiment above, for example... Figure 1 Steps S1 to S4 as described above.

[0125] Those skilled in the art will understand that the illustrations Figure 5 This is merely an example of a computer device and does not constitute a limitation on the computer device. It may include more or fewer components than shown, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0126] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the computer device, connecting various parts of the computer device via various interfaces and lines.

[0127] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0128] If the modules integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc.

[0129] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0130] Accordingly, embodiments of the present invention provide a computer-readable storage medium, the computer-readable storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the computer-readable storage medium is located to perform steps in the device reliability assessment as described in the above embodiments, for example... Figure 1 Steps S1 to S4 as described above.

[0131] In summary, the device reliability assessment method, system, device, and medium provided in this application address the technical problem of improving the accuracy of device reliability assessment. The method includes: acquiring an initial real-time time series dataset that reflects the state of the device under test, and preprocessing and standardizing the initial real-time time series dataset to obtain a target real-time time series dataset; extracting potential key features from the target real-time time series dataset using a trained deep autoencoder to obtain potential key features; extracting temporal features from the target real-time time series dataset using a trained self-attention mechanism encoder to obtain temporal features; fusing the potential key features and temporal features to obtain a first fused feature; reconstructing the first fused feature using a trained decoder to obtain reconstructed samples, calculating the reconstruction error of the reconstructed samples, and fusing the reconstruction error and the first fused feature to obtain a second fused feature; inputting the second fused feature into a trained Gaussian mixture model to obtain a real-time negative log-likelihood value, comparing the real-time negative log-likelihood value with a preset threshold to obtain a deviation value; and evaluating the device under test using a preset reliability assessment algorithm based on the deviation value to obtain the reliability assessment result of the device under test. The equipment reliability assessment method disclosed in this application combines a deep autoencoder with a self-attention mechanism encoder. This not only acquires high-order coupling features between time-series data but also obtains dependency and trend features, thereby introducing dynamic temporal features and improving the accuracy of subsequent equipment reliability assessments. It also introduces reconstruction errors to broaden the distinction between normal and abnormal samples, enhancing the ability of subsequent Gaussian mixture models to differentiate between them. Furthermore, it obtains the real-time negative log-likelihood value of the second fusion feature through the Gaussian mixture model, achieving higher accuracy and stronger explanatory power for anomaly detection. Finally, it performs equipment reliability assessment based on the deviation between the real-time negative log-likelihood value and a preset threshold, realizing dynamic assessment and prediction of equipment reliability and improving the accuracy of equipment reliability assessments.

[0132] The various embodiments in this specification are described in a progressive manner. For directly identical or similar parts of the embodiments, refer to each other. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments. It should be noted that the technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as the combination of these technical features does not contradict each other, it should be considered within the scope of this specification.

[0133] The embodiments described above are merely preferred embodiments of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various improvements and substitutions without departing from the technical principles of this application, and these improvements and substitutions should also be considered within the scope of protection of this application. Therefore, the scope of protection of this patent application should be determined by the scope of the claims.

Claims

1. A method for evaluating equipment reliability, characterized in that, The method includes: An initial real-time time series dataset reflecting the status of the device under test is collected, and the initial real-time time series dataset is preprocessed and standardized to obtain a target real-time time series dataset. During the operation of the device under test, an initial real-time time series dataset of parameters related to the device status is collected. The parameters reflecting the device status include mechanical parameters, electrical parameters, process and environmental parameters, and operating status and control parameters. A trained deep autoencoder is used to extract potential key features from the target real-time time series dataset to obtain potential key features. A trained self-attention mechanism encoder is used to extract temporal features from the target real-time time series dataset to obtain temporal features. The potential key features and the temporal features are fused to obtain a first fused feature. The first fusion feature is reconstructed using the trained decoder to obtain a reconstructed sample, and the reconstruction error of the reconstructed sample is calculated. The reconstruction error is then fused with the first fusion feature to obtain a second fusion feature. The second fusion feature is input into the trained Gaussian mixture model to obtain the real-time negative log-likelihood value. The real-time negative log-likelihood value is compared with a preset threshold to obtain the deviation value. Based on the deviation value, a preset reliability evaluation algorithm is used to evaluate the reliability of the device under test and obtain the reliability evaluation result of the device under test. The self-attention mechanism encoder, after training, extracts temporal features from the target real-time time series dataset to obtain temporal features, including: Location codes are added to the real-time time series data of the target real-time time series dataset to obtain the location information of the real-time time series data; A trained self-attention mechanism encoder is used to extract temporal features from the real-time time series dataset of the target and the location information to obtain temporal features; The reliability assessment algorithm consists of a risk proportion function and a reliability function; The risk proportion function is expressed as follows: in, Indicates time, Indicates the deviation value. express Time deviation value The risk rate below Represents the benchmark risk function. Represents the fitting coefficient. Represented by natural constant An exponential function with base 0; The reliability function is expressed as follows: in, express Time deviation value The reliability of the equipment is as follows.

2. The equipment reliability assessment method as described in claim 1, characterized in that, The calculation of the reconstruction error of the reconstructed sample includes: The cosine similarity algorithm is used to calculate the cosine distance between the target real-time time series dataset and the reconstructed sample to obtain the cosine similarity error; The Euclidean distance algorithm is used to calculate the Euclidean distance between the target real-time time series dataset and the reconstructed sample to obtain the Euclidean distance error; The cosine similarity error and the Euclidean distance error are fused to obtain the reconstruction error.

3. The equipment reliability assessment method as described in claim 1, characterized in that, Obtaining the preset threshold includes: A comprehensive loss function is set for the constructed deep autoencoder, the self-attention mechanism encoder, the decoder, and the Gaussian mixture model; The first historical time series dataset of the device to be detected is collected. Based on the comprehensive loss function and the first historical time series dataset, the deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model are trained end-to-end to obtain the trained deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model. A second historical time series dataset of the normal state of the device to be detected is collected. Based on the second historical time series dataset, the deep autoencoder, the self-attention mechanism encoder, the decoder and the Gaussian mixture model are used to perform corresponding processing to obtain a set of negative log-likelihood values ​​of the normal state. The set of negative log-likelihood values ​​for the normal state is analyzed, and a preset threshold is obtained based on the analysis results.

4. The equipment reliability assessment method as described in claim 3, characterized in that, The comprehensive loss function is expressed as follows: in, Represents the loss function. This indicates the calculation of the mean square error. This represents the initial input dataset. This represents the reconstructed samples obtained based on the initial input dataset. This represents the input to the Gaussian mixture model. This represents the output of the Gaussian mixture model. This represents the output weights of the Gaussian mixture model. Represents a regular term. Indicates the weight of the regularization term. The first Gaussian mixture model obtained by fitting the model represents the... Gaussian sub-distribution, The input of the Gaussian mixture model represents the first... One characteristic, This represents the total number of Gaussian sub-distributions obtained by fitting the Gaussian mixture model. This represents the feature dimension of the input to the Gaussian mixture model. Indicates the first The first Gaussian sub-distribution's covariance matrix The diagonal elements corresponding to each feature.

5. The equipment reliability assessment method as described in claim 1, characterized in that, The process of preprocessing and standardizing the initial real-time time series dataset to obtain the target real-time time series dataset includes: The initial real-time time series dataset is subjected to correlation analysis using the maximum information coefficient, and the real-time time series dataset is filtered based on the correlation analysis results to obtain the filtered initial real-time time series dataset. The initial real-time time series dataset after filtering is normalized to obtain the target real-time time series dataset.

6. A device reliability assessment system, used to implement the device reliability assessment method according to any one of claims 1-5, characterized in that, The system includes: a data preprocessing unit, a feature extraction unit, a reconstruction error fusion unit, and a reliability assessment unit; The data preprocessing unit is used to collect an initial real-time time series dataset that reflects the status of the device to be detected, and to preprocess and standardize the initial real-time time series dataset to obtain the target real-time time series dataset. The feature extraction unit is used to extract potential key features from the target real-time time series dataset using a trained deep autoencoder to obtain potential key features, to extract temporal features from the target real-time time series dataset using a trained self-attention mechanism encoder to obtain temporal features, and to fuse the potential key features and the temporal features to obtain a first fused feature. The reconstruction error fusion unit is used to reconstruct the first fusion feature using the trained decoder to obtain a reconstructed sample, calculate the reconstruction error of the reconstructed sample, and fuse the reconstruction error with the first fusion feature to obtain a second fusion feature. The reliability assessment unit is used to input the second fused feature into the trained Gaussian mixture model to obtain a real-time negative log-likelihood value, compare the real-time negative log-likelihood value with a preset threshold to obtain a deviation value, and evaluate the reliability of the device under test using a preset reliability assessment algorithm based on the deviation value to obtain the reliability assessment result of the device under test. The reliability assessment algorithm is set to consist of a risk proportion function and a reliability function.

7. A computer device, characterized in that: The computer device includes a memory, a processor, and a transceiver connected to each other via a bus; the memory stores a set of computer program instructions and data, and transmits the stored data to the processor, which executes the computer program instructions stored in the memory to perform the device reliability assessment method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program that, when executed, implements the device reliability assessment method as described in any one of claims 1 to 5.

Images