Machine learning-based device defect prediction method and system

Abstract

The present invention relates to the technical field of machine learning, and provides a machine learning-based device defect prediction method and system. The method comprises: collecting sample defect data of devices in chemical water-medium based operating, and performing classification and labeling processing on the sample defect data on the basis of defect types of the devices to obtain label parameter data; preprocessing the label parameter data, and on the basis of device defect characteristics, performing feature extraction on parameter data corresponding to different labels; on the basis of features extracted for different labels and the preprocessed label parameter data, constructing prediction functions for the corresponding labels by using a machine learning algorithm; and acquiring a validation data set of an actual production environment and inputting same into a defect prediction model constructed on the basis of the prediction functions for all the labels, and predicting device defects and feeding same back to the defect prediction model for model optimization and device defect assessment. Machine learning is used to perform prediction analysis on device defects, thereby improving prediction accuracy.

Description

A Machine Learning-Based Method and System for Predicting Equipment Defects Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a machine learning-based method and system for predicting equipment defects. Background Technology

[0002] Currently, machine learning methods for predicting equipment defects have made progress in technological development. However, due to the diversity of equipment defect types and characteristics, errors in the training data caused by machine learning training often result in low accuracy of the final defect type analysis results and poor interpretability of the model.

[0003] Therefore, this invention proposes a machine learning-based method and system for predicting equipment defects. Summary of the Invention

[0004] This invention provides a machine learning-based method and system for predicting equipment defects. It collects defect data from equipment, trains an iterative prediction model using machine learning on the defect data, and finally evaluates the prediction model and issues defect alerts based on real-time data. This method accurately locates the defects, improving the accuracy of equipment defect prediction.

[0005] On one hand, the present invention provides a machine learning-based method for predicting equipment defects, comprising:

[0006] Step 1: Collect sample defect data of equipment in chemical water operation, and classify and label the sample defect data according to the defect type of the equipment to obtain label parameter data;

[0007] Step 2: Preprocess the label parameter data and extract features from the parameter data under different labels based on the characteristics of equipment defects;

[0008] Step 3: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label;

[0009] Step 4: Obtain the verification dataset of the actual production environment and input it into the defect prediction model constructed by the prediction function under all labels to predict equipment defects and feed them back into the defect prediction model for model optimization and equipment defect assessment.

[0010] On the other hand, in step 1, sample defect data of equipment used in the chemical water operation are collected.

[0011] include:

[0012] Based on the type and location of the equipment in the chemical water operation, a first number is assigned to the corresponding equipment and a second number is assigned to the target sensor of the corresponding equipment.

[0013] Match the first and second numbers according to the sensor-equipment type lookup table in the installation guide to complete the monitoring of the operating data of the equipment in the chemical water operation and obtain sample defect data.

[0014] On the other hand, in step 1, the sample defect data is categorized based on the type of defect.

[0015] Class tag processing yields tag parameter data, including:

[0016] Keyword extraction was performed on all sample defect data according to the defect name under the equipment defect type;

[0017] Based on semantic similarity, the extracted keywords are clustered by synonyms to obtain several defect clusters. A classification label is then assigned to each defect cluster to obtain label parameter data. Here, one defect cluster corresponds to one classification label, and the number of defect clusters is less than the number of defect types.

[0018] On the other hand, in step 2, the label parameter data is preprocessed, including:

[0019] Obtain the standard parameter data corresponding to each label parameter data, and construct the first matrix A;

[0020] ;

[0021] in, This represents the standard parameter data corresponding to the i-th tag parameter data at the j-th monitoring time, where, , ;

[0022] Perform matrix standardization on matrix A to obtain the standardized matrix. .

[0023] On the other hand, in step 2, feature extraction is performed on the parameter data under different labels based on the characteristics of equipment defects, including:

[0024] Based on the type-feature mapping table, obtain the equipment defect features corresponding to the defect types of the equipment;

[0025] Based on the standardized matrix Calculate the covariance matrix and perform eigenvalue decomposition on the covariance matrix to obtain the standardized matrix. Component characteristics ;

[0026] according to Determine the component characteristics The contribution level of each feature element, where, m1 represents the f-th feature element; m1 represents the component feature. The number of characteristic elements present in the data; This indicates the contribution level of the f-th feature element;

[0027] From all feature elements, elements whose contribution is greater than a preset level are selected to form the main feature, and the defect feature function of the device is determined:

[0028] ;

[0029] in, The device represents the first Defect feature function with l feature elements Indicates the filtered first l feature elements, This indicates the characteristics of the equipment defect corresponding to the type of defect in the equipment. Represents the logarithmic function. Represents the transformation coefficients for the feature elements; This represents the conversion factor for the characteristics of equipment defects;

[0030] The defect characteristics of the device are determined based on all defect characteristic functions.

[0031] On the other hand, in step 3, the prediction function under the corresponding label is constructed, including:

[0032] The extracted features are compared with the preprocessed label parameter data to obtain a feature-data comparison list for each extracted feature, and the comparison degree of the corresponding extracted feature is calculated according to the feature-data comparison list.

[0033] ;

[0034] in, This represents the correlation coefficient of the d-th extracted feature; This represents the number of preprocessed label parameter data determined from the feature-data comparison list of the d-th extracted feature; This indicates the amount of data corresponding to the i2th preprocessed label parameter; This represents the allowable weight of the i2th preprocessed label parameter data for the dth extracted feature; This indicates the utilization rate of the d-th extracted feature using the corresponding i2-th preprocessed label parameter data;

[0035] Machine learning is performed on each feature-data reference list based on machine learning algorithms, and a prediction function is constructed according to the reference degree and the learning rate of the machine learning algorithm.

[0036] On the other hand, in step 4, predicting equipment defects and feeding them back into the defect prediction model for model optimization and equipment defect assessment includes:

[0037] Real-time parameter data of the equipment is obtained from the actual production environment. After preprocessing and label classification of the real-time parameter data, a verification dataset is obtained, wherein the verification dataset contains sub-datasets under different label classifications.

[0038] The verification dataset is input into the defect prediction model constructed by all prediction functions to obtain the first prediction array under each category label, wherein the first prediction array is obtained by performing defect prediction analysis on the verification dataset based on the defect prediction model;

[0039] Meanwhile, each subset of data is input into the prediction function under the corresponding classification label to obtain the corresponding second prediction array. The second prediction array is obtained by performing defect prediction analysis on the corresponding subset of data based on the prediction function under the corresponding classification label.

[0040] The first prediction array, the second prediction array, and the actual defect array under the same category label are presented in the same coordinate system, and three curves are plotted.

[0041] Obtain the defect arrays for the same defect feature from the three curves, and calculate the first defect difference, second defect difference, third defect difference, and first defect variance in the defect arrays respectively. Assign defect coefficients to the corresponding defect arrays. ;

[0042] ;

[0043] in, This corresponds to the first defect difference in the defect array. This corresponds to the second defect difference in the defect array; This corresponds to the third defect difference in the defect array; Based on Calculate the variance of the first defect;

[0044] The three curves are truncated according to the standard that the defect coefficient is greater than a preset coefficient, and the second defect variance of the truncated curve is determined. ;

[0045] ;

[0046] in, To extract the variance of the i3rd first defect involved in the curve; To extract the variance of all first-defect variances involved in the curve;

[0047] Obtain the time difference array of occurrence of adjacent defect features in the three curves. The third defect variance is determined, and then the fourth defect variance of the cut curve is calculated. Here, t01, t11 and t21 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the first defect feature in the adjacent defect features, respectively; t02, t12 and t22 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the second defect feature in the adjacent defect features, respectively.

[0048] Obtain the first area above the third curve corresponding to the first prediction array in the three curves and the second area below the third curve corresponding to the actual defect array.

[0049] At the same time, obtain the third area above the third curve corresponding to the second prediction array of the three curves and the fourth area below the third curve corresponding to the actual defect array.

[0050] The fifth defect variance is obtained based on the first ratio of the first area to the third area, the second ratio of the second area to the fourth area, the first overlap ratio of the first area to the third area, and the second overlap ratio of the second area to the fourth area.

[0051] Based on the second defect variance, the fourth defect variance, and the fifth defect variance, obtain the function optimization criteria of the prediction function under the corresponding classification label from the variance-label-optimization mapping table;

[0052] Input all the function optimization criteria into the standard analysis model in sequence to obtain the optimization vector;

[0053] The defect prediction model is optimized according to the optimization vector, and the equipment defect is evaluated on the validation dataset according to the optimized model.

[0054] On the other hand, including:

[0055] Data collection module: Collects sample defect data of equipment in chemical water operation, and classifies and labels the sample defect data according to the defect type to obtain label parameter data;

[0056] Feature parameter module: preprocesses the label parameter data and extracts features from the parameter data under different labels based on the characteristics of equipment defects;

[0057] Prediction function module: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label;

[0058] Evaluation module: Obtains the verification dataset of the actual production environment and inputs it into the defect prediction model constructed by the prediction function under all labels, predicts equipment defects and feeds them back into the defect prediction model for model optimization and equipment defect evaluation.

[0059] This invention provides a machine learning-based method and system for predicting equipment defects. It collects defect data from equipment, trains an iterative prediction model using machine learning on the defect data, and finally evaluates the prediction model and issues defect alerts based on real-time data. This method accurately locates the defects, improving the accuracy of equipment defect prediction. Attached Figure Description

[0060] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0061] Figure 1 is a flowchart illustrating a machine learning-based device defect prediction method provided in an embodiment of the present invention.

[0062] Figure 2 is a schematic diagram of the structure of a machine learning-based equipment defect prediction system provided in an embodiment of the present invention. Embodiments of the present invention

[0063] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0064] Example 1:

[0065] As shown in Figure 1, an embodiment of the present invention provides a machine learning-based equipment defect prediction method, comprising:

[0066] Step 1: Collect sample defect data of equipment in chemical water operation, and classify and label the sample defect data according to the defect type of the equipment to obtain label parameter data;

[0067] Step 2: Preprocess the label parameter data and extract features from the parameter data under different labels based on the characteristics of equipment defects;

[0068] Step 3: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label;

[0069] Step 4: Obtain the verification dataset of the actual production environment and input it into the defect prediction model constructed by the prediction function under all labels to predict equipment defects and feed them back into the defect prediction model for model optimization and equipment defect assessment.

[0070] In this embodiment, water operation in chemistry refers to the process of using water as a carrier medium for reactions, mass transfer, separation, and other operations during chemical industrial production.

[0071] In this embodiment, the sample defect data is a record of defect information collected from the chemical water operation equipment, including a detailed description of equipment failures, damage, abnormalities, and other problems, as well as relevant information about the equipment.

[0072] In this embodiment, the defect type refers to various problems or defect types that may occur in chemical water operation equipment, such as equipment failure, leakage, wear, blockage, corrosion, etc.

[0073] In this embodiment, the classification labeling process involves classifying and labeling the sample data.

[0074] In this embodiment, the label parameter data refers to the data obtained after classifying and labeling the collected sample defect data.

[0075] In this embodiment, preprocessing refers to a series of cleaning, transformation, and processing operations performed on the raw data before performing machine learning tasks.

[0076] In this embodiment, the characteristics of equipment defects refer to the features exhibited by different equipment defects in the sample data, including abnormal data such as the changing trends and ranges of parameters such as temperature, pressure, and flow rate.

[0077] In this embodiment, feature extraction refers to selecting and extracting the information that best describes the characteristics of the data from the original data as new features for machine learning tasks.

[0078] In this embodiment, machine learning algorithms are algorithms that utilize data and statistical techniques to enable a system to learn from data and continuously improve its performance. These include supervised learning algorithms, unsupervised learning algorithms, reinforcement learning algorithms, and semi-supervised learning algorithms.

[0079] In this embodiment, the prediction function is a function obtained by training a model through machine learning, which is used to map input data to predicted output results.

[0080] In this embodiment, the actual production environment refers to a real industrial production environment where equipment is running and generating data.

[0081] In this embodiment, the verification dataset is a data sample collected from the actual production environment, including data under normal operating conditions and data with known defects.

[0082] In this embodiment, the defect prediction model is a model used to predict future defects in the device.

[0083] In this embodiment, model optimization is the process of continuously adjusting and improving the parameters of a machine learning model to improve the model's performance and accuracy on real data.

[0084] In this embodiment, equipment defect assessment refers to the process of comprehensively evaluating and analyzing the equipment defect situation predicted by the prediction model.

[0085] The working principle and beneficial effects of the above technical solution are as follows: by collecting and processing equipment defect data, constructing a prediction model and optimizing the evaluation, accurate prediction and monitoring of defects in chemical water operation equipment can be achieved, problems can be detected in advance, production risks can be reduced, the location of defects can be accurately located, and the accuracy of equipment defect prediction can be improved.

[0086] Example 2:

[0087] Based on the above embodiment 1, step 1 involves collecting sample defect data of equipment used in the chemical water operation, including:

[0088] Based on the type and location of the equipment in the chemical water operation, a first number is assigned to the corresponding equipment and a second number is assigned to the target sensor of the corresponding equipment.

[0089] Match the first and second numbers according to the sensor-equipment type lookup table in the installation guide to complete the monitoring of the operating data of the equipment in the chemical water operation and obtain sample defect data.

[0090] In this embodiment, the equipment types include various water treatment equipment, such as water pumps, filters, heaters, reactors, etc.

[0091] In this embodiment, the device location refers to the specific installation location of the device in the chemical water treatment system, such as the water supply inlet, water outlet, circulation pipe, etc.

[0092] In this embodiment, the first number refers to the unique identification number of the corresponding device, which is used to distinguish and identify different devices in the monitoring system.

[0093] In this embodiment, the second number is a unique identification number pointing to the target sensor configuration of the corresponding device, which is used to distinguish and identify different sensors in the monitoring system.

[0094] In this embodiment, the sensor-device type lookup table is a lookup table that represents the mapping relationship between sensor types and device types.

[0095] The working principle and beneficial effects of the above technical solution are as follows: by configuring a unique number and sensor, the monitoring and defect data acquisition of the chemical water operation equipment are realized, which provides a data basis for accurately locating the defect and improves the accuracy of equipment defect prediction.

[0096] Example 3:

[0097] Based on the above embodiment 1, in step 1, the sample defect data is classified and labeled according to the defect type to obtain label parameter data, including:

[0098] Keyword extraction was performed on all sample defect data according to the defect name under the equipment defect type;

[0099] Based on semantic similarity, the extracted keywords are clustered by synonyms to obtain several defect clusters. A classification label is then assigned to each defect cluster to obtain label parameter data. Here, one defect cluster corresponds to one classification label, and the number of defect clusters is less than the number of defect types.

[0100] In this embodiment, the defect name is a classification and naming of defects that occur during the operation of the equipment, such as: leakage, blockage, wear, oxidation, etc.

[0101] In this embodiment, keyword extraction refers to extracting important words or phrases describing defects from sample defect data.

[0102] In this embodiment, synonym clustering refers to the process of grouping semantically similar words or phrases into the same cluster according to certain rules.

[0103] In this embodiment, defect clustering refers to the process of grouping similar defects into the same category.

[0104] The working principle and beneficial effects of the above technical solution are as follows: through steps such as keyword extraction, synonym clustering and configuration of classification labels, defect data can be better organized and managed, providing a data foundation for accurately locating the defect and improving the accuracy of equipment defect prediction.

[0105] Example 4:

[0106] Based on the above embodiment 1, step 2 involves preprocessing the tag parameter data, including:

[0107] Obtain the standard parameter data corresponding to each label parameter data, and construct the first matrix A;

[0108] ;

[0109] in, This represents the standard parameter data corresponding to the i-th tag parameter data at the j-th monitoring time, where, , ;

[0110] Perform matrix standardization on matrix A to obtain the standardized matrix. .

[0111] In this embodiment, the standard parameter data are the parameter values ​​when the device is operating normally under certain specific conditions.

[0112] In this embodiment, the first matrix is ​​a matrix composed of label parameter data and corresponding standard parameter data.

[0113] In this embodiment, matrix standardization refers to transforming the data in the matrix according to certain rules so that the data meets the standard.

[0114] The working principle and beneficial effects of the above technical solution are as follows: by acquiring standard parameter data, constructing matrix A, and performing standardization processing, data comparison and analysis are realized, providing a data basis for accurately locating the defect and improving the accuracy of equipment defect prediction.

[0115] Example 5:

[0116] Based on the above embodiment 4, in step 2, feature extraction is performed on the parameter data under different tags according to the characteristics of equipment defects, including:

[0117] Based on the type-feature mapping table, obtain the equipment defect features corresponding to the defect types of the equipment;

[0118] Based on the standardized matrix Calculate the covariance matrix and perform eigenvalue decomposition on the covariance matrix to obtain the standardized matrix. Component characteristics ;

[0119] according to Determine the component characteristics The contribution level of each feature element, where, m1 represents the f-th feature element; m1 represents the component feature. The number of characteristic elements present in the data; This indicates the contribution level of the f-th feature element;

[0120] From all feature elements, elements whose contribution is greater than a preset level are selected to form the main feature, and the defect feature function of the device is determined:

[0121] ;

[0122] in, The device represents the first Defect feature function with l feature elements Indicates the filtered first l feature elements, This indicates the characteristics of the equipment defect corresponding to the type of defect in the equipment. Represents the logarithmic function. Represents the transformation coefficients for the feature elements; This represents the conversion factor for the characteristics of equipment defects;

[0123] The defect characteristics of the device are determined based on all defect characteristic functions.

[0124] In this embodiment, the type-feature mapping table is a mapping table that represents the relationship between the types of defects in the device and the corresponding defect features for each type of defect.

[0125] In this embodiment, the characteristics of equipment defects are defects that occur during the operation or use of the equipment, including various physical characteristics, performance parameters, etc.

[0126] In this embodiment, the covariance matrix is ​​a matrix that describes the strength and direction of the relationship between defect types and defect characteristics.

[0127] In this embodiment, eigenvalue decomposition is used to decompose the covariance matrix into a set of eigenvectors and corresponding eigenvalues.

[0128] In this embodiment, component features refer to the new features obtained by eigenvalue decomposition of the covariance matrix.

[0129] In this embodiment, the degree of contribution is used as an indicator to measure the degree of contribution of each feature element in the component features to the overall data change.

[0130] In this embodiment, the principal feature refers to the component feature selected in the eigenvalue decomposition whose degree of sharing is greater than a preset degree based on the degree of contribution.

[0131] In this embodiment, the feature element refers to each element in the component features obtained by eigenvalue decomposition of the standardized matrix.

[0132] In this embodiment, the conversion coefficient is a coefficient used to convert feature elements and equipment defect characteristics.

[0133] The working principle and beneficial effects of the above technical solution are as follows: By decomposing the covariance matrix and analyzing its contribution, combined with the characteristics of equipment defects and transformation coefficients, a functional model that better reflects the features of equipment defects is constructed. This provides a data foundation for accurately locating defects and improves the accuracy of equipment defect prediction.

[0134] Example 6:

[0135] Based on the above embodiment 1, step 3 involves constructing a prediction function for the corresponding label, including:

[0136] The extracted features are compared with the preprocessed label parameter data to obtain a feature-data comparison list for each extracted feature, and the comparison degree of the corresponding extracted feature is calculated according to the feature-data comparison list.

[0137] ;

[0138] in, This represents the correlation coefficient of the d-th extracted feature; This represents the number of preprocessed label parameter data determined from the feature-data comparison list of the d-th extracted feature; This indicates the amount of data corresponding to the i2th preprocessed label parameter; This represents the allowable weight of the i2th preprocessed label parameter data for the dth extracted feature; This indicates the utilization rate of the d-th extracted feature using the corresponding i2-th preprocessed label parameter data;

[0139] Machine learning is performed on each feature-data reference list based on machine learning algorithms, and a prediction function is constructed according to the reference degree and the learning rate of the machine learning algorithm.

[0140] In this embodiment, the comparative analysis is a process of analyzing the interaction relationships between variables to represent the connections between them.

[0141] In this embodiment, the feature-data comparison list is a list compiled by comparing and organizing the extracted features with the preprocessed label parameter data, recording the correlation between each feature and the corresponding data.

[0142] In this embodiment, the correlation degree refers to the measure of the degree of association between a feature and data in the feature-data correlation list, representing the degree of association between each extracted feature and the corresponding data.

[0143] In this embodiment, the weight is allowed to be a coefficient representing the degree of influence of the feature on the label data.

[0144] In this embodiment, the learning rate refers to the hyperparameter used in machine learning algorithms to control the step size of model parameter updates.

[0145] The working principle and beneficial effects of the above technical solution are as follows: by analyzing the relationship between features and corresponding data through a feature-data comparison list, calculating the comparison degree, and applying machine learning algorithms to construct a prediction function, combined with learning rate optimization model training, a data foundation is provided for accurately locating the defect, thereby improving the accuracy of equipment defect prediction.

[0146] Example 7:

[0147] Based on the above embodiment 4, step 4, predicting equipment defects and feeding them back into the defect prediction model for model optimization and equipment defect assessment, includes:

[0148] Real-time parameter data of the equipment is obtained from the actual production environment. After preprocessing and label classification of the real-time parameter data, a verification dataset is obtained, wherein the verification dataset contains sub-datasets under different label classifications.

[0149] The verification dataset is input into the defect prediction model constructed by all prediction functions to obtain the first prediction array under each category label, wherein the first prediction array is obtained by performing defect prediction analysis on the verification dataset based on the defect prediction model;

[0150] Meanwhile, each subset of data is input into the prediction function under the corresponding classification label to obtain the corresponding second prediction array. The second prediction array is obtained by performing defect prediction analysis on the corresponding subset of data based on the prediction function under the corresponding classification label.

[0151] The first prediction array, the second prediction array, and the actual defect array under the same category label are presented in the same coordinate system, and three curves are plotted.

[0152] Obtain the defect arrays for the same defect feature from the three curves, and calculate the first defect difference, second defect difference, third defect difference, and first defect variance in the defect arrays respectively. Assign defect coefficients to the corresponding defect arrays. ;

[0153] ;

[0154] in, This corresponds to the first defect difference in the defect array. This corresponds to the second defect difference in the defect array; This corresponds to the third defect difference in the defect array; Based on Calculate the variance of the first defect;

[0155] The three curves are truncated according to the standard that the defect coefficient is greater than a preset coefficient, and the second defect variance of the truncated curve is determined. ;

[0156] ;

[0157] in, To extract the variance of the i3rd first defect involved in the curve; To extract the variance of all first-defect variances involved in the curve;

[0158] Obtain the time difference array of occurrence of adjacent defect features in the three curves. The third defect variance is determined, and then the fourth defect variance of the cut curve is calculated. Here, t01, t11 and t21 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the first defect feature in the adjacent defect features, respectively; t02, t12 and t22 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the second defect feature in the adjacent defect features, respectively.

[0159] Obtain the first area above the third curve corresponding to the first prediction array in the three curves and the second area below the third curve corresponding to the actual defect array.

[0160] At the same time, obtain the third area above the third curve corresponding to the second prediction array of the three curves and the fourth area below the third curve corresponding to the actual defect array.

[0161] The fifth defect variance is obtained based on the first ratio of the first area to the third area, the second ratio of the second area to the fourth area, the first overlap ratio of the first area to the third area, and the second overlap ratio of the second area to the fourth area.

[0162] Based on the second defect variance, the fourth defect variance, and the fifth defect variance, obtain the function optimization criteria of the prediction function under the corresponding classification label from the variance-label-optimization mapping table;

[0163] Input all the function optimization criteria into the standard analysis model in sequence to obtain the optimization vector;

[0164] The defect prediction model is optimized according to the optimization vector, and the equipment defect is evaluated on the validation dataset according to the optimized model.

[0165] In this embodiment, the real-time parameter data refers to various parameter data collected in real time during the operation of the device, such as temperature, pressure, humidity, and current.

[0166] In this embodiment, the subset dataset is a small dataset extracted from the validation dataset based on different label classifications.

[0167] In this embodiment, the first prediction array is the result of defect prediction analysis on the validation dataset based on the defect prediction model. It includes the predicted equipment defects for each category label.

[0168] In this embodiment, defect prediction analysis refers to the process of predicting and evaluating the defects of the equipment by analyzing and processing the real-time parameter data of the equipment and using a prediction model.

[0169] In this embodiment, the second prediction array is a set of prediction results obtained by performing defect prediction analysis on the corresponding subset of data based on the prediction function under the corresponding classification label.

[0170] In this embodiment, the defect array refers to the data obtained by analyzing the verification dataset based on the defect prediction model after the real-time parameter data of the device has been preprocessed and labeled.

[0171] In this embodiment, the first defect difference refers to the defect array under the same defect feature after the first predicted array, the second predicted array, and the actual defect array under the same classification label are presented in the same coordinate system.

[0172] In this embodiment, the second defect difference refers to the difference between the second prediction array obtained by the prediction function under each classification label and the actual defect array after the defect prediction model analyzes the validation dataset.

[0173] In this embodiment, the first prediction array, the second prediction array, and the actual defect array are presented in the same coordinate system to draw three curves, and the difference between the third curve and the other arrays is shown.

[0174] In this embodiment, the defect coefficient is a quantitative evaluation index of equipment defects, which measures the severity of defects under different defect characteristics.

[0175] In this embodiment, the first defect variance is a difference measure calculated after analyzing the first prediction array, the second prediction array, and the actual defect array under the same classification label.

[0176] In this embodiment, the second defect variance is determined by comparing the differences between the second prediction array and the actual defect array corresponding to different prediction functions under the same classification label in the same coordinate system.

[0177] In this embodiment, the third defect variance is determined by comparing the differences between the third prediction array and the actual defect array corresponding to different prediction functions under the same classification label in the same coordinate system.

[0178] In this embodiment, the fourth defect variance is obtained by analyzing and comparing three different curves, combined with the variance of the cut curve obtained by the third defect variance.

[0179] In this embodiment, the first area refers to the area above the third curve corresponding to the first prediction array.

[0180] In this embodiment, the second area refers to the area below the third curve corresponding to the actual defect array.

[0181] In this embodiment, the third area is the area above the third curve corresponding to the second prediction array in the three curves, which is located above the third curve corresponding to the actual defect array.

[0182] In this embodiment, the area below the third curve corresponding to the fourth area second prediction data.

[0183] In this embodiment, the first ratio refers to the ratio of the first area to the third area.

[0184] In this embodiment, the second ratio refers to the ratio of the second area to the fourth area.

[0185] In this embodiment, the first overlap ratio refers to the ratio of the area of ​​the first curve corresponding to the first prediction array above the third curve corresponding to the actual defect array to the total area above the third curve corresponding to the actual defect array, i.e., first area / (first area + third area).

[0186] In this embodiment, the second overlap ratio refers to the second area / (second area + fourth area).

[0187] In this embodiment, the fifth defect variance is a defect variance calculated based on the first ratio of the first area to the third area, the second ratio of the second area to the fourth area, the first overlap ratio of the first area to the third area, and the second overlap ratio of the second area to the fourth area.

[0188] In this embodiment, the variance-label-optimization mapping table is a table that records the relationship between variance and label, as well as the corresponding optimization criteria.

[0189] In this embodiment, the function optimization criterion refers to the optimization criterion for the corresponding variance and label data found under the variance-label-optimization mapping table based on the label classification.

[0190] In this embodiment, the standard analysis model is a specific model used to determine the criteria for function optimization.

[0191] In this embodiment, the optimization vector refers to a set of vectors calculated based on all optimization criteria and the standard analysis model, which is used to optimize the defect prediction model.

[0192] The working principle and beneficial effects of the above technical solution are as follows: by acquiring the parameter data of the equipment in real time, and by using the defect prediction model and prediction function to predict and optimize defects, a data foundation is provided for accurately locating the defect, thereby improving the accuracy of equipment defect prediction.

[0193] Example 8:

[0194] As shown in Figure 2, an embodiment of the present invention provides a machine learning-based equipment defect prediction system, characterized in that it includes:

[0195] Data collection module: Collects sample defect data of equipment in chemical water operation, and classifies and labels the sample defect data according to the defect type to obtain label parameter data;

[0196] Feature parameter module: preprocesses the label parameter data and extracts features from the parameter data under different labels based on the characteristics of equipment defects;

[0197] Prediction function module: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label;

[0198] Evaluation module: Obtains the verification dataset of the actual production environment and inputs it into the defect prediction model constructed by the prediction function under all labels, predicts equipment defects and feeds them back into the defect prediction model for model optimization and equipment defect evaluation.

[0199] The working principle and beneficial effects of the above technical solution are as follows: by collecting and processing equipment defect data, constructing a prediction model and optimizing the evaluation, accurate prediction and monitoring of defects in chemical water operation equipment can be achieved, problems can be detected in advance, production risks can be reduced, the location of defects can be accurately located, and the accuracy of equipment defect prediction can be improved.

[0200] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A machine learning-based method for predicting equipment defects, characterized in that, include: Step 1: Collect sample defect data of equipment in chemical water operation, and classify and label the sample defect data according to the defect type of the equipment to obtain label parameter data; Step 2: Preprocess the label parameter data and extract features from the parameter data under different labels based on the characteristics of equipment defects; Step 3: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label; Step 4: Obtain the verification dataset of the actual production environment and input it into the defect prediction model constructed by the prediction function under all labels to predict equipment defects and feed them back into the defect prediction model for model optimization and equipment defect assessment.

2. The machine learning-based equipment defect prediction method according to claim 1, characterized in that, In step 1, sample defect data of equipment used in chemical water operation are collected, including: Based on the type and location of the equipment in the chemical water operation, a first number is assigned to the corresponding equipment and a second number is assigned to the target sensor of the corresponding equipment. Match the first and second numbers according to the sensor-equipment type lookup table in the installation guide to complete the monitoring of the operating data of the equipment in the chemical water operation and obtain sample defect data.

3. The machine learning-based equipment defect prediction method according to claim 1, characterized in that, In step 1, the sample defect data is classified and labeled based on the defect type to obtain label parameter data, including: Keyword extraction was performed on all sample defect data according to the defect name under the defect type of the equipment. Based on semantic similarity, the extracted keywords are clustered into synonyms to obtain several defect clusters. A classification label is then assigned to each defect cluster to obtain label parameter data. Each defect cluster corresponds to one classification label, and the number of defect clusters is less than the number of defect types.

4. The machine learning-based equipment defect prediction method according to claim 1, characterized in that, Step 2 involves preprocessing the tag parameter data, including: Obtain the standard parameter data corresponding to each label parameter data, and construct the first matrix A; ； in, This represents the standard parameter data corresponding to the i-th tag parameter data at the j-th monitoring time, where, ， ； Perform matrix standardization on matrix A to obtain the standardized matrix. 。 5. The machine learning-based equipment defect prediction method according to claim 4, characterized in that, In step 2, feature extraction is performed on the parameter data under different tags based on the characteristics of equipment defects, including: Based on the type-feature mapping table, obtain the equipment defect features corresponding to the defect types of the equipment; Based on the standardized matrix Calculate the covariance matrix and perform eigenvalue decomposition on the covariance matrix to obtain the standardized matrix. Component characteristics ； according to Determine the component characteristics The contribution level of each feature element, where, m1 represents the f-th feature element; m1 represents the component feature. The number of characteristic elements present in the data; This indicates the contribution level of the f-th feature element; From all feature elements, elements whose contribution is greater than a preset level are selected to form the main feature, and the defect feature function of the device is determined: ； in, The device represents the first Defect feature function with l feature elements Indicates the filtered first l feature elements, This indicates the characteristics of the equipment defect corresponding to the type of defect in the equipment. Represents the logarithmic function. Represents the transformation coefficients for the feature elements; This represents the conversion factor for the characteristics of equipment defects; The defect characteristics of the device are determined based on all defect characteristic functions.

6. The machine learning-based equipment defect prediction method according to claim 1, characterized in that, In step 3, the prediction function for the corresponding label is constructed, including: The extracted features are compared with the preprocessed label parameter data to obtain a feature-data comparison list for each extracted feature, and the comparison degree of the corresponding extracted feature is calculated according to the feature-data comparison list. ； in, This represents the correlation coefficient of the d-th extracted feature; This represents the number of preprocessed label parameter data determined from the feature-data comparison list of the d-th extracted feature; This indicates the amount of data corresponding to the i2th preprocessed label parameter; This represents the allowable weight of the i2th preprocessed label parameter data for the dth extracted feature; This indicates the utilization rate of the d-th extracted feature using the corresponding i2-th preprocessed label parameter data; Machine learning is performed on each feature-data reference list based on machine learning algorithms, and a prediction function is constructed according to the reference degree and the learning rate of the machine learning algorithm.

7. The machine learning-based equipment defect prediction method according to claim 1, characterized in that, In step 4, equipment defects are predicted and fed back into the defect prediction model for model optimization and equipment defect assessment, including: Real-time parameter data of the equipment is obtained from the actual production environment. After preprocessing and label classification of the real-time parameter data, a verification dataset is obtained, wherein the verification dataset contains sub-datasets under different label classifications. The verification dataset is input into the defect prediction model constructed by all prediction functions to obtain the first prediction array under each category label, wherein the first prediction array is obtained by performing defect prediction analysis on the verification dataset based on the defect prediction model; Meanwhile, each subset of data is input into the prediction function under the corresponding classification label to obtain the corresponding second prediction array. The second prediction array is obtained by performing defect prediction analysis on the corresponding subset of data based on the prediction function under the corresponding classification label. The first prediction array, the second prediction array, and the actual defect array under the same category label are presented in the same coordinate system, and three curves are plotted. Obtain the defect arrays for the same defect feature from the three curves, and calculate the first defect difference, second defect difference, third defect difference, and first defect variance in the defect arrays respectively. Assign defect coefficients to the corresponding defect arrays. ； ； in, This corresponds to the first defect difference in the defect array. This corresponds to the second defect difference in the defect array; This corresponds to the third defect difference in the defect array; For based on Calculate the variance of the first defect; The three curves are truncated according to the standard that the defect coefficient is greater than a preset coefficient, and the second defect variance of the truncated curve is determined. ； ； in, To extract the variance of the i3rd first defect involved in the curve; To extract the variance of all first-defect variances involved in the curve; Obtain the time difference array of occurrence of adjacent defect features in the three curves. The third defect variance is determined, and then the fourth defect variance of the cut curve is calculated. Here, t01, t11 and t21 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the first defect feature in the adjacent defect features, respectively; t02, t12 and t22 are the occurrence times of the first predicted defect, the second predicted defect and the actual defect under the second defect feature in the adjacent defect features, respectively. Obtain the first area above the third curve corresponding to the first prediction array in the three curves and the second area below the third curve corresponding to the actual defect array. At the same time, obtain the third area above the third curve corresponding to the second prediction array of the three curves and the fourth area below the third curve corresponding to the actual defect array. The fifth defect variance is obtained based on the first ratio of the first area to the third area, the second ratio of the second area to the fourth area, the first overlap ratio of the first area to the third area, and the second overlap ratio of the second area to the fourth area. Based on the second defect variance, the fourth defect variance, and the fifth defect variance, obtain the function optimization criteria of the prediction function under the corresponding classification label from the variance-label-optimization mapping table; Input all the function optimization criteria into the standard analysis model in sequence to obtain the optimization vector; The defect prediction model is optimized according to the optimization vector, and the equipment defect is evaluated on the validation dataset according to the optimized model.

8. A machine learning-based equipment defect prediction system, characterized in that, include: Data collection module: Collects sample defect data of equipment in chemical water operation, and classifies and labels the sample defect data according to the defect type to obtain label parameter data; Feature parameter module: preprocesses the label parameter data and extracts features from the parameter data under different labels based on the characteristics of equipment defects; Prediction function module: Based on the features extracted under different labels and the preprocessed label parameter data, and combined with machine learning algorithms, construct the prediction function for the corresponding label; Evaluation module: Obtains the verification dataset of the actual production environment and inputs it into the defect prediction model constructed by the prediction function under all labels, predicts equipment defects and feeds them back into the defect prediction model for model optimization and equipment defect evaluation.

Images