A process fault separation method and device of adaptive Gaussian mixture model

Abstract

The application discloses a process fault separation method and device of an adaptive Gaussian mixture model. The features include: obtaining manufacturing process data, performing normalization processing on the manufacturing process data to determine processed process data; performing feature extraction on the processed process data by a target adaptive Gaussian mixture model to determine process feature data and reconstructed process data; performing fault contribution degree analysis based on the process feature data and the reconstructed process data to determine a target fault variable. The application can accurately analyze the multi-modal data distribution generated in an industrial manufacturing process, determine the contribution degree of each variable to a fault, timely locate a fault source and take countermeasures, so as to avoid unnecessary economic losses and casualties.

Description

Technical Field

[0001] This invention relates to the field of industrial process quality control technology, and in particular to a process fault separation method and apparatus using an adaptive Gaussian mixture model. Background Technology

[0002] As time progresses, actual industrial processes become increasingly complex, variable, and interconnected, posing new challenges to quality control in industrial manufacturing. In actual industrial processes, due to constant changes in the internal and external environment, processes need to operate under varying conditions. Therefore, process fault detection and fault variable isolation are crucial for ensuring product quality and smooth process operation. Currently, due to changes in market demand, fluctuations in raw material prices, and equipment aging, industrial processes operate at different points during their lifespan. This process exhibits multimodal characteristics, meaning that historical data has different distribution characteristics depending on specific operating specifications. How to adaptively fit multimodal distribution data and achieve fault isolation under multi-condition manufacturing is a pressing issue. In this context, traditional multivariate statistical process monitoring methods, such as principal component analysis and partial least squares, cannot be directly applied to multimodal process modeling because these methods assume that the data follows a unimodal distribution, an assumption that is often difficult to meet under dynamic multi-condition production conditions. Furthermore, most process control methods in manufacturing focus on fault detection, largely neglecting fault variable isolation. Summary of the Invention

[0003] This invention provides a process fault separation method and apparatus based on an adaptive Gaussian mixture model to solve the technical problem in the prior art of being unable to locate and separate fault variables in industrial manufacturing processes.

[0004] According to one aspect of the present invention, a process fault separation method based on an adaptive Gaussian mixture model is provided, comprising:

[0005] Acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data;

[0006] The process data is subjected to feature extraction using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training the process data on a pre-built adaptive Gaussian mixture model;

[0007] Based on the process characteristic data and the reconstruction process data, a fault contribution analysis is performed to determine the target fault variable.

[0008] According to another aspect of the present invention, a process fault separation device for an adaptive Gaussian mixture model is provided, comprising:

[0009] The preprocessing module is used to acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data.

[0010] The feature processing module is used to extract features from the processing data using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training the process data on a pre-built adaptive Gaussian mixture model;

[0011] The fault analysis module is used to perform fault contribution analysis based on the process characteristic data and the reconstruction process data to determine the target fault variable.

[0012] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores a computer program that can be executed by the at least one processor, which enables the at least one processor to perform the process fault separation method of the adaptive Gaussian mixture model according to any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the process fault separation method of the adaptive Gaussian mixture model according to any embodiment of the present invention.

[0017] The technical solution of this invention acquires manufacturing process data, normalizes the manufacturing process data to determine the processing process data, and normalization can effectively reduce data complexity and improve analysis accuracy. It then uses a target adaptive Gaussian mixture model to extract features from the processing process data to determine process feature data and reconstructed process data. Based on the process feature data and the reconstructed process data, it performs fault contribution analysis to determine target fault variables. This solves the technical problem in existing technologies where fault variables cannot be located and separated during industrial manufacturing. It enables accurate analysis of the multimodal data distribution generated during industrial manufacturing, determines the contribution of each variable to the fault, and promptly locates the source of the fault and takes countermeasures to avoid unnecessary economic losses and casualties.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A flowchart of a process fault separation method using an adaptive Gaussian mixture model is provided as an embodiment of the present invention;

[0021] Figure 2 A flowchart of a process fault separation method using an adaptive Gaussian mixture model provided in an embodiment of the present invention;

[0022] Figure 3 This is a schematic diagram of the structure of a process fault separation device based on an adaptive Gaussian mixture model provided in an embodiment of the present invention;

[0023] Figure 4 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Figure 1 This invention provides a flowchart of a process fault separation method using an adaptive Gaussian mixture model. This embodiment is applicable to fault identification and separation using multimodal data in multi-condition manufacturing scenarios. The method can be executed by a process fault separation device based on an adaptive Gaussian mixture model. This device can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0027] S110. Obtain manufacturing process data, normalize the manufacturing process data, and determine the processing process data.

[0028] Manufacturing process data can be multimodal data generated during industrial manufacturing processes, which can be used to monitor, control, optimize, and analyze manufacturing activities. It should be noted that manufacturing process data can consist of multiple different process variables; for example, the process variables in manufacturing process data can be reactor cooling water flow rate and reactor temperature.

[0029] Optionally, the processing data can be normalized manufacturing process data. It should be noted that, since the processing data, as process data during the online monitoring phase, is acquired sequentially, the normalization parameters are selected from the maximum and minimum values ​​of the offline training data to normalize the real-time acquired manufacturing process data. For example, through... These represent the i-th sample of the offline training data before and after normalization, respectively. Let J represent the j-th sample of manufacturing process data before and after normalization, and the j-th sample of data for determining the processing procedure, respectively. These represent the maximum and minimum values ​​of the offline training data, respectively. The normalization process is shown below:

[0030]

[0031] Specifically, manufacturing process data is acquired through online monitoring, and the manufacturing process data is normalized based on offline training data to determine the processing process data.

[0032] S120. The process data is processed by feature extraction using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data.

[0033] The target adaptive Gaussian mixture model is trained on a pre-built adaptive Gaussian mixture model using training process data. This training process data can be pre-collected manufacturing process data, labeled to distinguish between normal and faulty process data. It should be noted that the manufacturing process data in the training process data is typically historical manufacturing data from the same scenario.

[0034] Optionally, the adaptive Gaussian mixture model can be a pre-built neural network model. The target adaptive Gaussian mixture model consists of a variational autoencoder and a decoder. The target adaptive Gaussian mixture model can extract feature data of each process variable in the manufacturing process data and reconstruct the manufacturing process data through the decoder.

[0035] Optionally, the process feature data can be feature data obtained by the variational autoencoder of the target adaptive Gaussian mixture model to extract features from the process data; the reconstructed process data can be process data reconstructed by the encoder of the target adaptive Gaussian mixture model based on the process feature data.

[0036] Specifically, the process data is feature extracted using an adaptive Gaussian mixture model to determine process feature data and reconstruct process data.

[0037] Optionally, in this invention, the first adaptive Gaussian mixture model is determined by training an adaptive Gaussian mixture model pre-constructed based on training process data by maximizing the log-likelihood function.

[0038] Identify the number of mixture terms in the training Gaussian mixture term of the first adaptive Gaussian mixture model;

[0039] At least one updated Gaussian mixture term is determined based on the number of mixture terms;

[0040] The first adaptive Gaussian mixture model is updated based on the updated Gaussian mixture term to obtain the second adaptive Gaussian mixture model;

[0041] If the model acceptance probability of the second adaptive Gaussian mixture model meets the preset training objective, then the second adaptive Gaussian mixture model is determined as the target adaptive Gaussian mixture model.

[0042] Optionally, the first adaptive Gaussian mixture model can be an adaptive Gaussian mixture model trained based on the training process data. The adaptive Gaussian mixture model consists of a variational autoencoder and a decoder. The input training process data is the process data, i.e., the feature representations of normal process data and fault process data. The variational autoencoder performs dimensionality reduction and feature extraction on the feature representations of the input normal process data and fault process data, outputting latent features. Within the variational autoencoder, an adaptive Gaussian mixture unit splits and / or merges the Gaussian components of the latent features to enhance the adaptability of the Gaussian mixture term to changes in observed values. The decoder reconstructs the latent features into process data and iteratively converges to reconstruct the latent feature representation. That is, by setting the maximization of the log-likelihood function as the core optimization objective during training, model parameter learning and latent feature representation reconstruction are performed, and iterative convergence is conducted to obtain the first adaptive Gaussian mixture model. For example, the training process data is represented by x, and the latent feature representation is represented by z. The process of maximizing the log-likelihood function during training is as follows:

[0043]

[0044] in, Indicates decoder, Indicates encoder, and These represent the network parameters of the encoder and decoder, respectively, and DKL represents the Kullback-Leible divergence distance.

[0045] Optionally, the training Gaussian mixture term can be a Gaussian mixture term in the first adaptive Gaussian mixture model; the number of mixture terms can be the total number of training Gaussian mixture terms. It should be noted that, after training the first adaptive Gaussian mixture model, splitting and merging operations are introduced based on the number of mixture terms to enhance the adaptability of the Gaussian mixture term to changes in observations.

[0046] Optionally, the updated Gaussian mixture term can be obtained by splitting or merging the training Gaussian mixture term.

[0047] Optionally, the second adaptive Gaussian mixture model can be an adaptive Gaussian mixture model obtained by modifying the structure of the Gaussian mixture model by introducing an updated Gaussian mixture term.

[0048] Optionally, the model acceptance probability can be understood as an indicator data for evaluating the second adaptive Gaussian mixture model. It should be noted that when introducing the updated Gaussian mixture term, if the log-likelihood value based on maximizing the log-likelihood function is increased, the model acceptance probability is greater; if the log-likelihood value based on maximizing the log-likelihood function is not increased, the model acceptance probability is smaller.

[0049] Optionally, the preset training objective can be a condition used to evaluate the training results of the adaptive Gaussian mixture model. It should be noted that, in this invention, the preset training objective can be set to improve the second adaptive Gaussian mixture model that maximizes the log-likelihood function.

[0050] Optionally, in this invention, if the update of the first adaptive Gaussian mixture model based on the updated Gaussian mixture term obtained from the splitting and merging operations improves the output log-likelihood value of the model, the updated second adaptive Gaussian mixture model can be retained; if the update of the first adaptive Gaussian mixture model based on the updated Gaussian mixture term obtained from the splitting and merging operations does not improve the output log-likelihood value of the model, the updated second adaptive Gaussian mixture model is discarded.

[0051] Specifically, the first adaptive Gaussian mixture model is determined by training an adaptive Gaussian mixture model pre-constructed based on training process data by maximizing the log-likelihood function; the number of mixture terms in the training Gaussian mixture terms of the first adaptive Gaussian mixture model is identified; at least one updated Gaussian mixture term is determined based on the number of mixture terms; the first adaptive Gaussian mixture model is updated based on the updated Gaussian mixture term to obtain a second adaptive Gaussian mixture model; if the model acceptance probability of the second adaptive Gaussian mixture model meets the preset training objective, then the second adaptive Gaussian mixture model is determined as the target adaptive Gaussian mixture model.

[0052] Optionally, in another optional embodiment of the present invention, determining at least one updated Gaussian mixture term based on the number of mixture terms includes:

[0053] If the number of mixture terms is less than a preset mixture term threshold, then one of the training Gaussian mixture terms is split to obtain at least one updated Gaussian mixture term; if the number of mixture terms is greater than the preset mixture term threshold, then two of the training Gaussian mixture terms are merged to determine at least one updated Gaussian mixture term.

[0054] Optionally, the mixture threshold can be a preset quantity threshold. It should be noted that if the number of mixtures is less than the preset mixture threshold, a training Gaussian mixture term needs to be selected for the splitting operation; if the number of mixtures is greater than the preset mixture threshold, two Gaussian mixture terms are selected for the merging operation.

[0055] Specifically, the relationship between the number of mixture terms and the mixture term threshold is identified. If the number of mixture terms is less than the preset mixture term threshold, a training Gaussian mixture term is selected for splitting to obtain at least one updated Gaussian mixture term. If the number of mixture terms is greater than the preset mixture term threshold, two training Gaussian mixture terms are selected for merging to determine at least one updated Gaussian mixture term. For example, assuming the i-th training Gaussian mixture term is selected for splitting, the newly generated Gaussian mixture terms m and n can be defined as follows:

[0056]

[0057]

[0058]

[0059] in, Let m be the weight of the Gaussian mixture term m generated after splitting at the t-th iteration; Let n be the weight of the Gaussian mixture term n generated after splitting at the t-th iteration; Let m be the weight allocation ratio of component m in the t-th iteration; The weights of the split training Gaussian mixture; The weights of the split training Gaussian mixture; Let m be the mean vector of the Gaussian mixture term m after splitting; For the i-th training Gaussian mixture term, a subset of the data The mean; Let m be the covariance matrix of the split component m during iteration; This is the old covariance matrix at the (t-1)th iteration of the split; , and This represents the small perturbation vector used for the mean of the perturbation;

[0060] Suppose that the p-th and q-th Gaussian mixture terms are chosen for the merging operation, and the newly generated Gaussian mixture term can be represented as:

[0061]

[0062]

[0063]

[0064] in, Let h be the weight of the newly generated Gaussian component h after merging at the t-th iteration; The weight of the p-th Gaussian component being merged at the (t-1)-th iteration; The weight of the q-th Gaussian component being merged at the (t-1)-th iteration; Let h be the mean vector of the merged components h at the t-th iteration; Let be the mean vector of the p-th component at the (t-1)-th iteration; Let be the mean vector of the q-th component at the (t-1)-th iteration; Let h be the covariance matrix of the merged component h at the t-th iteration; Let be the covariance matrix of the p-th component at the (t-1)-th iteration; Let be the covariance matrix of the q-th component at the (t-1)-th iteration; is the transpose of the mean deviation, used to correct the covariance shift caused by changes in the mean after merging; T is used to denote the transpose of the matrix.

[0065] The model acceptance probability P is calculated to determine whether to retain the newly generated model. The model acceptance probability P can be expressed as:

[0066]

[0067] Among them, This represents the newly generated second adaptive Gaussian mixture model. This represents the first adaptive Gaussian mixture model.

[0068] S130. Based on the process characteristic data and the reconstruction process data, perform fault contribution analysis to determine the target fault variable.

[0069] Optionally, the target fault variable can be a process variable that causes a fault during the manufacturing process. It should be noted that after acquiring manufacturing process data, if a fault is detected in one of the manufacturing process data, since the manufacturing process data consists of multiple process variables, the target fault variable causing the fault is identified by analyzing each process variable. For example, in the Tennessee-Eastman industrial process, there are a total of 52 process variables, including 41 measured variables and 11 manipulated variables. This process has 21 different types of faults. The manufacturing process data was acquired at the 161st sampling point. A fault of type 14 would cause an impact on the reactor cooling water valves, which would affect the reactor cooling water flow rate and the reactor cooling outlet temperature. Through process variable analysis of the manufacturing process data obtained at the 161st sampling point, the reactor cooling water flow rate is identified as the target fault variable.

[0070] Specifically, fault contribution analysis is performed based on process characteristic data and reconstructed process data to determine target fault variables.

[0071] The technical solution of this invention acquires manufacturing process data, normalizes the manufacturing process data to determine the processing process data, and normalization can effectively reduce data complexity and improve analysis accuracy. It then uses a target adaptive Gaussian mixture model to extract features from the processing process data to determine process feature data and reconstructed process data. Based on the process feature data and the reconstructed process data, it performs fault contribution analysis to determine target fault variables. This solves the technical problem in existing technologies where fault variables cannot be located and separated during industrial manufacturing. It enables accurate analysis of the multimodal data distribution generated during industrial manufacturing, determines the contribution of each variable to the fault, and promptly locates the source of the fault and takes countermeasures to avoid unnecessary economic losses and casualties.

[0072] Figure 2 This is a flowchart illustrating a process fault separation method using an adaptive Gaussian mixture model, provided as an embodiment of the present invention. The relationship between this embodiment and the previous embodiments is that this specifically describes the process of identifying faults and separating process variables. Figure 2 As shown, the method includes:

[0073] S210. Obtain manufacturing process data, normalize the manufacturing process data, and determine the processing process data.

[0074] S220. The process data is subjected to feature extraction using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data.

[0075] S230. Calculate the Hotelling statistic and the squared prediction error based on the process feature data and the reconstructed process data.

[0076] Optionally, for the process feature data extracted from the manufacturing process data, the Hotelling statistic corresponding to each feature is calculated. For example, the Hotelling statistic is calculated using T... 2 To represent, h i The feature data representing the i-th manufacturing process data. and They represent h respectively i The mean and variance of the mean, and the calculation method of the Hotelling statistic is as follows:

[0077]

[0078] Optionally, for the reconstructed process data corresponding to the manufacturing process data, the squared prediction error corresponding to the manufacturing process data is calculated. For example, the squared prediction error of the i-th manufacturing process data is represented by SPE(i), where x... i This represents the data for the i-th manufacturing process. Let represent the reconstructed process data corresponding to the i-th manufacturing process data. The squared prediction error SPE(i) is calculated as follows:

[0079]

[0080] Specifically, the Hotelling statistic and squared prediction error are calculated based on process characteristic data and reconstructed process data.

[0081] S240. If the Hotling statistic and the squared prediction error satisfy the fault condition, then the manufacturing process data is considered to be fault process data.

[0082] Optionally, the fault conditions can be pre-set conditions used to assess whether faults exist in the process characteristic data. It should be noted that by analyzing the Hotelling statistic and squared prediction error of the process characteristic data, it is determined whether the Hotelling statistic and squared prediction error meet the fault conditions. If the fault conditions are met, the manufacturing process data is considered faulty and identified as faulty process data; if the fault conditions are not met, the manufacturing process data is considered not faulty, and fault identification and separation are performed on the next collected process fault data.

[0083] Specifically, by analyzing the Hotelling statistic and squared prediction error of the process characteristic data in relation to the fault conditions, if the Hotelling statistic and squared prediction error meet the fault conditions, the manufacturing process data is considered faulty process data; if the Hotelling statistic and squared prediction error do not meet the fault conditions, the next batch of collected manufacturing process data is analyzed.

[0084] Optionally, in another optional embodiment of the present invention, the step of considering the manufacturing process data as fault process data if the Hotelling statistic and the squared prediction error satisfy the fault condition includes:

[0085] Based on the target adaptive Gaussian mixture model and the training process data, a first control limit and a second control limit are determined; if the Hotling statistic exceeds the first control limit and the squared prediction error exceeds the second control limit, then the Hotling statistic and the squared prediction error are considered to meet the fault condition.

[0086] Optionally, the first control limit can be understood as the control limit corresponding to the Hotelling statistic of the training process data, and the confidence level of the first control limit is set to 0.99.

[0087] Optionally, the second control limit can be understood as the control limit corresponding to the squared prediction error of the training process data, and the confidence level of the second control limit is set to 0.99.

[0088] Optionally, the process characteristic data can be analyzed for faults using the first and second control limits. Typically, 99% of the process characteristic data will fall below the first and second control limits, meaning that this type of process characteristic data is not faulty. However, if 1% of the process characteristic data slightly exceeds the first and second control limits, it can be considered that this type of process characteristic data has random fluctuations, leading to controllable errors, and is also considered to be faultless. If a process characteristic data significantly exceeds the first and second control limits, and deviates from the overall pattern of normal process characteristic data, it indicates that the manufacturing process data is abnormal and is faulty process data.

[0089] Optionally, if the Hotelling statistic of the process characteristic data exceeds the first control limit, and the squared prediction error of the process characteristic data exceeds the second control limit, then the Hotelling statistic and the squared prediction error are considered to meet the fault conditions, and the process characteristic data is faulty process data. If the Hotelling statistic of the process characteristic data does not exceed the first control limit, and the squared prediction error of the process characteristic data does not exceed the second control limit, then the Hotelling statistic and the squared prediction error are considered not to meet the fault conditions, and the manufacturing process data is normal process data.

[0090] Specifically, the first and second control limits are determined based on the target adaptive Gaussian mixture model and training process data; if the Hotling statistic exceeds the first control limit, and the squared prediction error exceeds the second control limit, then the Hotling statistic and the squared prediction error are considered to meet the fault condition.

[0091] Optionally, in another optional embodiment of the present invention, determining the first control limit and the second control limit based on the target adaptive Gaussian mixture model and the training process data includes:

[0092] For each training process data, feature extraction and feature reconstruction are performed on the training process data through the target adaptive Gaussian mixture model to determine the training feature data and training reconstruction data.

[0093] For each training process data, the training Hotelling statistic and the training squared prediction error are calculated based on the training feature data and the training reconstruction data.

[0094] The first control limit and the second control limit are determined by calculating the control limits based on the training Hotelling statistic and the training squared prediction error corresponding to all the training feature data using the Gaussian kernel density estimation method.

[0095] Optionally, after training the target adaptive Gaussian mixture model, for the training process data, each training process data is sequentially input into the target adaptive Gaussian mixture model, and feature extraction and feature reconstruction are performed on each training process data to obtain the training feature data and training reconstruction data corresponding to each training process data.

[0096] Optionally, for each training process's feature data and reconstructed data, the training Hotelling statistic and training squared prediction error corresponding to each training process data are calculated. It should be noted that the calculation method for the training Hotelling statistic and training squared prediction error of the training process data is the same as in the above embodiment.

[0097] Optionally, after calculating the Hotelling statistic and training squared prediction error for each training process data, a first control limit for the Hotelling statistic and a second control limit corresponding to the training squared prediction error are calculated based on the training Hotelling statistic and training squared prediction error using the Gaussian kernel density estimation method, using the Gaussian kernel density estimation method. For example, the first control limit is calculated using... The second control limit is indicated by... The calculation method for the first control limit is as follows:

[0098]

[0099] Where δ represents the window width parameter, and N represents the total number of training data.

[0100] The second control limit is calculated as follows:

[0101]

[0102] Optionally, after calculating the first and second control limits, fault detection performance analysis is performed on the first and second control limits using the training process data and its labels. The performance of the first and second control limits is measured using two metrics: False Alarm Rate (FAR) and Fault Detection Rate (FDR). For example, T... N and T F This represents the number of normal training data points and the number of faulty training data points that exceeded the first or second control limit, expressed as N. N and N F These represent the number of data points during normal training and the number of data points during faulty training, respectively. The specific calculation method is as follows:

[0103]

[0104] If the false alarm rate and fault detection rate meet the detection performance requirements, the first and second control limits are used as the statistics for detecting faults; if the false alarm rate and fault detection rate do not meet the detection performance requirements, the first and second control limits are recalculated using the target adaptive Gaussian mixture model until the false alarm rate and fault detection rate meet the detection performance requirements.

[0105] S250. Perform fault contribution analysis on the fault process data to determine the target fault variable.

[0106] Optionally, after identifying faulty process data, a fault contribution analysis is performed on the process variables associated with the faulty process data to determine the target fault variable.

[0107] Optionally, in another optional embodiment of the present invention, the step of performing fault contribution analysis on the fault process data to determine the target fault variable includes:

[0108] For each process variable of the fault process data, a contribution analysis is performed based on the reconstruction loss of the process variable to determine the contribution of the process variable; the process variable corresponding to the highest contribution is taken as the target fault variable.

[0109] Optionally, the contribution of process variables can be used to represent the likelihood that a process variable will cause a failure. It should be noted that, since the failure process data consists of multiple process variables, the contribution of each process variable is calculated by analyzing the reconstruction loss of each process variable during feature extraction and reconstruction in the target adaptive Gaussian mixture model. For example, the contribution of a process variable is represented by CD(i,j), where i represents the i-th failure process data and j represents the j-th process variable of the i-th failure process data; through... This represents the input data of the j-th process variable in the i-th fault process data, through... This represents the reconstructed data of the j-th process variable in the i-th fault process data. The contribution of the process variable is calculated using CD(i,j) as follows:

[0110]

[0111] Optionally, after calculating the contribution of each process variable, the process variable with the highest contribution is selected as the target fault variable.

[0112] Specifically, for each process variable in the fault process data, a contribution analysis is performed based on the reconstruction loss of the process variable to determine its contribution. The process variable with the highest contribution is then selected as the target fault variable.

[0113] The embodiments of the present invention can accurately define the boundaries of process fault states, effectively improve the accuracy and reliability of fault detection, solve the technical problem in the prior art that fault variables cannot be located and separated in the industrial manufacturing process, accurately analyze the distribution of multimodal data generated in the industrial manufacturing process, effectively improve the accuracy and reliability of fault detection, locate the source of fault in time and take countermeasures to avoid unnecessary economic losses and casualties.

[0114] Figure 3 This is a schematic diagram of a process fault separation device based on an adaptive Gaussian mixture model, provided in an embodiment of the present invention. Figure 3 As shown, the device includes: a preprocessing module 310, a feature processing module 320, and a fault analysis module 330; wherein,

[0115] Preprocessing module 310 is used to acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data;

[0116] The feature processing module 320 is used to extract features from the processing data using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training the process data on a pre-built adaptive Gaussian mixture model;

[0117] The fault analysis module 330 is used to perform fault contribution analysis based on the process characteristic data and the reconstruction process data to determine the target fault variable.

[0118] The technical solution of this invention includes a preprocessing module that effectively reduces data complexity and improves analysis accuracy through normalization; a feature processing module that adaptively fits the complex multimodal distribution of process data, accurately extracts key process features, and achieves high-quality data reconstruction, effectively reducing data dimensionality; and a fault analysis module that accurately locates key variables causing faults and quantifies the impact of each variable on the fault, solving the technical problem in existing technologies where fault variables cannot be located and separated during industrial manufacturing. This allows for accurate analysis of the multimodal data distribution generated during industrial manufacturing, determining the contribution of each variable to the fault, timely locating the fault source, and taking countermeasures to avoid unnecessary economic losses and casualties.

[0119] Optionally, the fault analysis module 330 is specifically used for:

[0120] The Hotelling statistic and squared prediction error are calculated based on the process feature data and the reconstructed process data.

[0121] If the Hotling statistic and the squared prediction error satisfy the fault condition, then the manufacturing process data is considered to be fault process data.

[0122] The fault contribution analysis is performed on the fault process data to determine the target fault variable.

[0123] Optionally, the fault analysis module 330 is further used for:

[0124] The first and second control limits are determined based on the target adaptive Gaussian mixture model and the training process data.

[0125] If the Hotling statistic exceeds the first control limit, and the squared prediction error exceeds the second control limit, then the Hotling statistic and the squared prediction error are considered to meet the fault conditions.

[0126] Optionally, the fault analysis module 330 is further used for:

[0127] For each training process data, feature extraction and feature reconstruction are performed on the training process data through the target adaptive Gaussian mixture model to determine the training feature data and training reconstruction data.

[0128] For each training process data, the training Hotelling statistic and the training squared prediction error are calculated based on the training feature data and the training reconstruction data.

[0129] The first control limit and the second control limit are determined by calculating the control limits based on the training Hotelling statistic and the training squared prediction error corresponding to all the training feature data using the Gaussian kernel density estimation method.

[0130] Optionally, the device further includes: a model training module, a model analysis module, a model update module, and a model evaluation module;

[0131] The model training module is used to train an adaptive Gaussian mixture model pre-built based on the training process data by maximizing the log-likelihood function, and to determine the first adaptive Gaussian mixture model.

[0132] The model analysis module is used to identify the number of mixture terms in the training Gaussian mixture term of the first adaptive Gaussian mixture model, and determine at least one updated Gaussian mixture term based on the number of mixture terms.

[0133] The model update module is used to update the first adaptive Gaussian mixture model based on the updated Gaussian mixture term to obtain the second adaptive Gaussian mixture model.

[0134] The model evaluation module is used to determine the second adaptive Gaussian mixture model as the target adaptive Gaussian mixture model if the model acceptance probability of the second adaptive Gaussian mixture model meets the preset training objective.

[0135] Optionally, the model analysis module is specifically used to: if the number of mixture terms is less than a preset mixture term threshold, select one of the training Gaussian mixture terms for splitting to obtain at least one updated Gaussian mixture term; if the number of mixture terms is greater than the preset mixture term threshold, select two of the training Gaussian mixture terms for merging to determine at least one updated Gaussian mixture term.

[0136] Optionally, the fault analysis module 330 is further used for:

[0137] For each process variable of the fault process data, a contribution analysis is performed based on the reconstruction loss of the process variable to determine the contribution of the process variable;

[0138] The process variable with the highest contribution is taken as the target fault variable.

[0139] The process fault separation device for adaptive Gaussian mixture model provided in this embodiment of the invention can execute the process fault separation method for adaptive Gaussian mixture model provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0140] Figure 4 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0141] like Figure 4 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0142] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of monitors, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer grids such as the Internet and / or various telecommunications grids.

[0143] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the process fault isolation method using an adaptive Gaussian mixture model.

[0144] In some embodiments, the adaptive Gaussian mixture model process fault isolation method can be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the adaptive Gaussian mixture model process fault isolation method described above can be performed. Alternatively, in other embodiments, processor 11 can be configured to perform the adaptive Gaussian mixture model process fault isolation method by any other suitable means (e.g., by means of firmware).

[0145] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0146] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0147] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0148] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0149] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or grid browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication grid). Examples of communication grids include local area networks (LANs), wide area networks (WANs), blockchain grids, and the Internet.

[0150] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS servers, such as high management difficulty and weak business scalability.

[0151] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0152] This embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the process fault separation method steps of the adaptive Gaussian mixture model as provided in any embodiment of the present invention. The method includes:

[0153] Acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data;

[0154] The process data is subjected to feature extraction using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training a pre-built adaptive Gaussian mixture model using the process data.

[0155] Based on the process characteristic data and the reconstruction process data, a fault contribution analysis is performed to determine the target fault variable.

[0156] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0157] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0158] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0159] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of mesh, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0160] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a grid of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0161] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0162] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A process fault separation method based on an adaptive Gaussian mixture model, characterized in that, include: Acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data; The process data is subjected to feature extraction using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training a pre-built adaptive Gaussian mixture model using the process data. Based on the process characteristic data and the reconstruction process data, a fault contribution analysis is performed to determine the target fault variable; Before acquiring manufacturing process data, normalizing the manufacturing process data, and determining the processed process data, the method further includes: The first adaptive Gaussian mixture model is determined by training an adaptive Gaussian mixture model pre-constructed based on training process data by maximizing the log-likelihood function. Identify the number of mixture terms in the training Gaussian mixture term of the first adaptive Gaussian mixture model, and determine at least one updated Gaussian mixture term based on the number of mixture terms; The first adaptive Gaussian mixture model is updated based on the updated Gaussian mixture term to obtain the second adaptive Gaussian mixture model; If the model acceptance probability of the second adaptive Gaussian mixture model meets the preset training objective, then the second adaptive Gaussian mixture model is determined as the target adaptive Gaussian mixture model. The step of determining at least one updated Gaussian mixture term based on the number of mixture terms includes: If the number of mixture terms is less than a preset mixture term threshold, then select one of the training Gaussian mixture terms for splitting to obtain at least one updated Gaussian mixture term; If the number of mixture terms is greater than a preset mixture term threshold, then two of the training Gaussian mixture terms are selected and merged to determine at least one updated Gaussian mixture term.

2. The method according to claim 1, characterized in that, The step of performing fault contribution analysis based on the process characteristic data and the reconstructed process data to determine the target fault variable includes: The Hotelling statistic and squared prediction error are calculated based on the process feature data and the reconstructed process data. If the Hotling statistic and the squared prediction error satisfy the fault condition, then the manufacturing process data is considered to be fault process data. The fault contribution analysis is performed on the fault process data to determine the target fault variable.

3. The method according to claim 2, characterized in that, The statement that if the Hotling statistic and the squared prediction error satisfy the fault condition, the manufacturing process data is considered fault process data includes: The first and second control limits are determined based on the target adaptive Gaussian mixture model and the training process data. If the Hotling statistic exceeds the first control limit, and the squared prediction error exceeds the second control limit, then the Hotling statistic and the squared prediction error are considered to meet the fault conditions.

4. The method according to claim 3, characterized in that, Determining the first and second control limits based on the target adaptive Gaussian mixture model and the training process data includes: For each training process data, feature extraction and feature reconstruction are performed on the training process data through the target adaptive Gaussian mixture model to determine the training feature data and training reconstruction data. For each training process data, the training Hotelling statistic and the training squared prediction error are calculated based on the training feature data and the training reconstruction data. The first control limit and the second control limit are determined by calculating the control limits based on the training Hotelling statistic and the training squared prediction error corresponding to all the training feature data using the Gaussian kernel density estimation method.

5. The method according to claim 2, characterized in that, The step of performing fault contribution analysis on the fault process data to determine the target fault variable includes: For each process variable of the fault process data, a contribution analysis is performed based on the reconstruction loss of the process variable to determine the contribution of the process variable; The process variable with the highest contribution is taken as the target fault variable.

6. A process fault separation device for an adaptive Gaussian mixture model, characterized in that, include: The preprocessing module is used to acquire manufacturing process data, normalize the manufacturing process data, and determine the processing process data. The feature processing module is used to extract features from the processing data using a target adaptive Gaussian mixture model to determine process feature data and reconstruct process data; wherein, the target adaptive Gaussian mixture model is obtained by training the process data on a pre-built adaptive Gaussian mixture model; The fault analysis module is used to perform fault contribution analysis based on the process characteristic data and the reconstruction process data to determine the target fault variable; The device further includes: a model training module, a model analysis module, a model update module, and a model evaluation module; The model training module is used to train an adaptive Gaussian mixture model pre-constructed based on training process data by maximizing the log-likelihood function, and to determine the first adaptive Gaussian mixture model. The model analysis module is used to identify the number of mixture terms in the training Gaussian mixture term of the first adaptive Gaussian mixture model, and determine at least one updated Gaussian mixture term based on the number of mixture terms. The model update module is used to update the first adaptive Gaussian mixture model based on the updated Gaussian mixture term to obtain the second adaptive Gaussian mixture model. The model evaluation module is used to determine the second adaptive Gaussian mixture model as the target adaptive Gaussian mixture model if the model acceptance probability of the second adaptive Gaussian mixture model meets the preset training objective. The model analysis module is specifically used for: if the number of mixture terms is less than a preset mixture term threshold, then selecting one of the training Gaussian mixture terms for splitting to obtain at least one updated Gaussian mixture term; if the number of mixture terms is greater than the preset mixture term threshold, then selecting two of the training Gaussian mixture terms for merging to determine at least one updated Gaussian mixture term.

7. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the process fault separation method of the adaptive Gaussian mixture model as described in any one of claims 1-5.

8. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the process fault separation method of any one of claims 1-5 using the adaptive Gaussian mixture model.

Images