Industrial automatic motor detection method and system

Abstract

The invention belongs to the technical field of motor detection, and discloses an industrial automatic motor detection method and system, and the method comprises the steps: obtaining motor mechanical performance data, motor electrical performance data and motor environment data; de-noising the motor mechanical performance data and the motor electrical performance data by adopting a dynamic error constraint recursive filtering method to obtain a filter output data set, and generating a pre-processing data set; aiming at the preprocessed data set, constructing a multi-modal driving enhanced regression technology, and predicting motor load data; performing space-time association based on the motor load data and the motor environment data, and constructing a space-time association diagram; the time-space association graph is embedded into a multi-criterion Gaussian detection method to calculate the comprehensive probability density of data points; a standard threshold value is set, and when the probability density of the real-time data points deviates from the standard threshold value, a grading early warning mechanism is triggered; the accuracy and reliability of motor detection can be improved, and an efficient solution is provided for real-time monitoring and maintenance of an industrial automatic motor.

Description

Technical Field

[0001] The present invention relates to the technical field of motor detection, and in particular to an industrial automation motor detection method and system. Background Art

[0002] Existing motor detection methods suffer from numerous problems. First, most rely on a single monitoring method or data source, making it difficult to comprehensively assess the motor's condition. These methods focus solely on electrical data such as current and voltage, neglecting mechanical performance or environmental factors. This single-minded approach can lead to early signs of motor anomalies not being detected in a timely manner, making it difficult to provide accurate assessments.

[0003] Existing motor detection methods typically rely on simple data acquisition and processing methods, which are susceptible to noise and abnormal data. In this case, the processed data may be biased, resulting in incorrect motor operation detection;

[0004] Existing motor fault detection methods often use simple rules to judge, which makes it difficult to cope with the complex changes and abnormal conditions in motor operation. Especially when the motor load and environment change drastically, simple threshold settings often cannot accurately identify faults and are prone to false alarms or missed alarms.

[0005] In view of this, the present invention proposes an industrial automation motor detection method and system to solve the above problems. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art and achieve the above-mentioned objectives, the present invention provides the following technical solution, a method for detecting an industrial automation motor, comprising:

[0007] Step A1: Acquire motor mechanical performance data, motor electrical performance data, and motor environmental data; wherein comprehensive temperature data is generated by fusing infrared thermal imaging with embedded sensors;

[0008] Step A2: Denoising the motor mechanical performance data and the motor electrical performance data using a dynamic error constrained recursive filtering method to obtain a filter output data set, and generating a preprocessed data set with a unified time and space reference; wherein the dynamic error constrained recursive filtering method obtains the filter output data set by continuously updating the filter coefficients and the error covariance matrix;

[0009] Step A3: Constructing a multimodal drive enhanced regression technique for the preprocessed dataset, performing feature selection on the preprocessed dataset by iteratively optimizing the sample weight distribution, and predicting the motor load data; the multimodal drive enhanced regression technique optimizes the sample weights by iteratively training weak learners and weighting sample errors, and ultimately constructing a strong learner through a weighted combination of multiple weak learners to predict the motor load data;

[0010] Step A4: Perform spatiotemporal correlation based on motor load data and motor environment data to construct a spatiotemporal correlation graph; embed the spatiotemporal correlation graph into a multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points; and set a standard threshold. When the probability density of real-time data points deviates from the standard threshold, a graded warning mechanism is triggered; in the multi-criteria Gaussian detection method, the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

[0011] Furthermore, the motor mechanical performance data includes vibration data and speed data;

[0012] Motor electrical performance data includes current data, comprehensive temperature data, voltage data and power data;

[0013] The motor environment data includes corrosive gas concentration data, electromagnetic intensity data, ambient temperature data, ambient humidity data, and ambient noise data.

[0014] Furthermore, the method for obtaining the comprehensive temperature data includes:

[0015] Use an infrared thermal imager to obtain a thermal image of the motor surface. Use OpenCV to process the thermal image. Extract the RGB value of each pixel from the thermal image and convert it into temperature data through color mapping. Store the temperature data of each pixel in a temperature matrix. Get the maximum temperature value in the temperature matrix and define it as the maximum temperature data of the motor surface.

[0016] Use embedded temperature sensors to collect stator winding temperature data, rotor core temperature data, bearing temperature data, and cooling medium temperature data inside the motor;

[0017] The maximum temperature data of the motor surface, the stator winding temperature data, the rotor core temperature data, the bearing temperature data and the cooling medium temperature data are weighted averaged to obtain the comprehensive temperature data.

[0018] Furthermore, the method for obtaining the preprocessed data set includes:

[0019] Based on the acquired motor mechanical performance data and motor electrical performance data, the interpolation method is used to handle missing values, and the Z-Score method is used to handle outliers to obtain a preliminary data set;

[0020] Based on the preliminary data set, the dynamic error constrained recursive filtering method is used to remove noise and obtain the filter output data set;

[0021] The filter output data set is then normalized and the timestamps are unified to obtain the preprocessed data set.

[0022] Furthermore, the specific method of using the dynamic error constrained recursive filtering method to remove noise and obtain the filter output data set includes:

[0023] Step o1: take the preliminary dataset as input;

[0024] Step o2: Initialize the parameters of the dynamic error constrained recursive filtering method, including filter coefficients, forgetting factors, and error covariance matrix;

[0025] Step o3, calculating the gain data at the nth moment by using a covariance weighted gain solution method based on the preliminary data set at the nth moment, the transpose of the preliminary data set at the nth moment, the error covariance matrix at the n-1th moment, and the forgetting factor;

[0026] Update the error covariance matrix based on the gain data and the error covariance matrix at time n-1;

[0027] calculating a filter output data set using the filter coefficients and the preliminary data set;

[0028] Calculate error data using the expected data set and the filter output data set; wherein the expected data set, i.e., reference data without noise, is obtained by consulting relevant materials;

[0029] updating filter coefficients according to the gain data and the error data;

[0030] Step o4: Repeat the above updating steps, set the error threshold, and when the error data is less than the set error threshold, stop the iteration and obtain the filter output data set.

[0031] Furthermore, the specific method of using the multimodal drive enhanced regression technology to predict motor load data includes:

[0032] Step M1: The input sample set includes h_h samples, and each sample in the input sample set includes a preprocessed data set and corresponding motor load data; the data in the preprocessed data set is the feature variable, and the motor load data is the target variable;

[0033] Step M2: Set the maximum number of iterations and initialize the sample weights;

[0034] Step M3, iteratively training the weak learner: In the tth iteration, the current sample weight is used to train the weak regressor to obtain the motor load prediction data predicted by the weak learner, where the weak learner is a linear regression model;

[0035] Based on the difference between the motor load prediction data and the motor load data predicted by the weak learner for the sample in the t-th iteration, combined with the corresponding sample weights, the weighted error of the weak learner in the current round is obtained by the weighted error method;

[0036] The weighted error of the weak learner is used to calculate the weight of the weak learner;

[0037] Based on the weighted error of the weak learner, update the weight of each sample;

[0038] Step M4: Repeat step M3 until the maximum number of iterations is reached and then stop iterating;

[0039] Step M5: constructing a strong learner by weighting the motor load prediction data predicted by the weak learner, that is, the motor load data finally outputted.

[0040] Furthermore, the specific method of constructing the spatiotemporal correlation graph includes:

[0041] The node system of the spatiotemporal correlation graph consists of physical entity nodes and virtual causal nodes. The physical entity nodes include motor load data and motor environment data. The virtual causal nodes include load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data.

[0042] Among them, load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data are calculated by Granger causality analysis method;

[0043] The nodes in the acquired node system are timestamped uniformly, the node system with the same timestamp is extracted, and each data in the node system is normalized; the weights between the object entity nodes and the virtual causal nodes in the node system are calculated using the ST-CPF method;

[0044] The specific method of ST-CPF is as follows:

[0045] Use matplotlib to draw a scatter plot to determine whether the relationship between nodes is linear or nonlinear; if it is linear, use the Pearson correlation method to calculate the correlation between nodes; if it is nonlinear, use the mutual information method to calculate the correlation between nodes; generate a correlation matrix;

[0046] Use Euclidean distance to calculate the distance between nodes; generate a distance matrix;

[0047] Based on the correlation matrix and its proximity matrix, the weights between nodes are calculated using the linear weighting method;

[0048] A spatiotemporal correlation graph is formed based on the nodes, the weights between nodes, and the corresponding timestamps.

[0049] Furthermore, the specific method of embedding the spatiotemporal correlation graph into the multi-criteria Gaussian detection method to calculate the comprehensive probability density includes:

[0050] Step V1: Input motor load prediction data and motor environment data, and perform missing value, outlier, normalization and unified timestamp processing to obtain preprocessed motor load prediction data and motor environment data;

[0051] Step V2, initialize the mean, covariance matrix and mixing weight parameters of the Gaussian component;

[0052] Step V3, using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components;

[0053] Step V4: Calculate the probability density of each Gaussian component, and use the probability density of the Gaussian component to calculate the responsibility of the Gaussian component;

[0054] Step V5: updating the mean, covariance matrix and mixing weight parameters of the Gaussian component according to the degree of responsibility;

[0055] Step V6: Repeat steps V4 and V5 until the set number of iterations is reached and the update stops;

[0056] Step V7: Calculate the comprehensive probability density of each data point.

[0057] Furthermore, the specific method of using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components includes:

[0058] According to the empirical rule, the candidate value range of the number of Gaussian components is set. For each candidate number of Gaussian components, the log-likelihood value of the multi-criteria Gaussian detection method is calculated. The AIC value and BIC value are calculated using the log-likelihood value. The AIC value and BIC value are weighted averaged to obtain a comprehensive index. By comparing the comprehensive indexes of different candidate numbers of Gaussian components, the candidate value with the smallest comprehensive index is selected as the optimal number of Gaussian components.

[0059] For each data point, the probability density of each candidate Gaussian component of each data point is calculated; then, the probability density of the comprehensive candidate Gaussian component of each data point is calculated by weighted averaging based on the mixing weight corresponding to the probability density of each Gaussian component; based on the natural logarithm method, the probability densities of the comprehensive candidate Gaussian components of N data points are added to obtain the log-likelihood value;

[0060] The AIC value is calculated based on twice the number of parameters of the Gaussian mixture model and twice the log-likelihood value;

[0061] The BIC value is calculated by multiplying the number of parameters and the log-likelihood value of the multi-criteria Gaussian detection method by the overall log-likelihood value times 2;

[0062] Among them, the number of parameters of the multi-criteria Gaussian detection method is obtained by multiplying the number of mean parameters, the number of mixing weight parameters, and the number of covariance matrix parameters constructed by the number of mean parameters by the number of candidate Gaussian components.

[0063] Furthermore, the specific method of setting the standard threshold and detecting whether the operation of the motor is abnormal by comparing the comprehensive probability density with the standard threshold includes:

[0064] The initial threshold is set by the historical comprehensive probability density; the responsibility and mean of the Gaussian component are used to calculate the historical reconstructed data points; the reconstruction error between the historical data points and the historical reconstructed data points is obtained based on the Euclidean distance; and the average value of the reconstruction error is used to correct the initial threshold to obtain the standard threshold.

[0065] A new set of data points is input and a new comprehensive probability density is obtained through the Gaussian mixture model. When the new comprehensive probability density is greater than or equal to the standard threshold, the motor operates normally; when the new comprehensive probability density is less than the standard threshold, the motor operates abnormally and an early warning is issued.

[0066] An industrial automation motor detection system is implemented based on the aforementioned industrial automation motor detection method, comprising:

[0067] Multi-source data acquisition module: acquires motor mechanical performance data, motor electrical performance data, and motor environmental data; it generates comprehensive temperature data by fusing infrared thermal imaging with embedded sensors;

[0068] Dynamic filtering processing module: This module uses a dynamic error-constrained recursive filtering method to denoise the motor mechanical and electrical performance data, obtain a filter output dataset, and generate a preprocessed dataset with a unified time and space reference. The dynamic error-constrained recursive filtering method obtains the filter output dataset by continuously updating the filter coefficients and error covariance matrix.

[0069] Motor Load Prediction Module: A multimodal drive-enhanced regression technique is constructed for preprocessed datasets. Feature selection is performed on the preprocessed dataset by iteratively optimizing the sample weight distribution, and motor load data is predicted. This technique optimizes sample weights by iteratively training weak learners and weighting sample errors. Finally, a strong learner is constructed through a weighted combination of multiple weak learners to predict motor load data.

[0070] Spatiotemporal correlation multi-criteria early warning module: Based on the spatiotemporal correlation of motor load data and motor environment data, a spatiotemporal correlation graph is constructed; the spatiotemporal correlation graph is embedded in the multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points; and a standard threshold is set. When the probability density of real-time data points deviates from the standard threshold, a graded early warning mechanism is triggered; in the multi-criteria Gaussian detection method, the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

[0071] The technical effects and advantages of the industrial automation motor detection method and system of the present invention are as follows:

[0072] The present invention can comprehensively combine the motor's mechanical performance, electrical performance, and motor environmental data to perform comprehensive preprocessing, aiming to improve the motor's operating stability. It also uses a dynamic error-constrained recursive filtering method to remove noise, thereby ensuring the accuracy of the processed data and laying a solid foundation for subsequent motor load prediction and operating status diagnosis.

[0073] Using multimodal drive-enhanced regression technology to predict motor load data can effectively improve the accuracy of the prediction model. This is because each round of weak learners optimizes the error of the previous round, gradually reducing the model's deviation and providing more accurate load predictions, thereby effectively assisting in motor operating status assessment and early warning.

[0074] In terms of anomaly detection, a multi-criteria Gaussian detection method is adopted. This method determines whether the motor has an abnormality by calculating the comprehensive probability density of motor load data and motor environment data. The AIC and BIC criteria are used to optimize the number of Gaussian components to ensure the rationality and accuracy of the method structure. In addition, the initial threshold is adjusted by calculating the reconstruction error between the data point and the historical reconstructed data point to obtain the standard threshold, which makes the motor anomaly detection more sensitive and robust. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a schematic diagram of an industrial automation motor detection method of the present invention;

[0076] Figure 2 Schematic diagram of the structure of the multi-criteria Gaussian detection method of the present invention;

[0077] Figure 3 This is a schematic diagram of an industrial automation motor detection system according to the present invention. DETAILED DESCRIPTION

[0078] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0079] Example 1, please refer to Figure 1 As shown, the industrial automation motor detection method described in this embodiment includes:

[0080] Step A1: Acquire motor mechanical performance data, motor electrical performance data, and motor environmental data; wherein comprehensive temperature data is generated by fusing infrared thermal imaging with embedded sensors;

[0081] Motor mechanical performance data includes vibration data and speed data;

[0082] Motor electrical performance data includes current data, comprehensive temperature data, voltage data and power data;

[0083] Motor environmental data includes corrosive gas concentration data, electromagnetic intensity data, ambient temperature data, ambient humidity data, and ambient noise data;

[0084] Vibration and speed data are obtained by installing vibration sensors and Hall effect sensors on the motor bearings; current, voltage, and power data are obtained by installing smart meters on the motor junction box; and corrosive gas concentration data, electromagnetic intensity data, ambient temperature data, ambient humidity data, and noise data are obtained by installing gas sensors, electromagnetic field sensors, temperature sensors, humidity sensors, and noise sensors in the motor's environment.

[0085] Step A2: Denoising the motor mechanical performance data and the motor electrical performance data using a dynamic error constrained recursive filtering method to obtain a filter output data set, and generating a preprocessed data set with a unified time and space reference; wherein the dynamic error constrained recursive filtering method obtains the filter output data set by continuously updating the filter coefficients and the error covariance matrix;

[0086] Step A3: Constructing a multimodal drive enhanced regression technique for the preprocessed dataset, performing feature selection on the preprocessed dataset by iteratively optimizing the sample weight distribution, and predicting the motor load data; the multimodal drive enhanced regression technique optimizes the sample weights by iteratively training weak learners and weighting sample errors, and ultimately constructing a strong learner through a weighted combination of multiple weak learners to predict the motor load data;

[0087] Step A4: Perform spatiotemporal correlation based on motor load data and motor environment data to construct a spatiotemporal correlation graph; embed the spatiotemporal correlation graph into a multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points; and set a standard threshold. When the probability density of real-time data points deviates from the standard threshold, a graded warning mechanism is triggered; in the multi-criteria Gaussian detection method, the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

[0088] Methods for obtaining comprehensive temperature data include:

[0089] Use an infrared thermal imager to obtain a thermal image of the motor surface. Use OpenCV to process the thermal image. Extract the RGB value of each pixel from the thermal image and convert it into temperature data through color mapping. Store the temperature data of each pixel in a temperature matrix. Get the maximum temperature value in the temperature matrix and define it as the maximum temperature data of the motor surface.

[0090] Use embedded temperature sensors to collect stator winding temperature data, rotor core temperature data, bearing temperature data, and cooling medium temperature data inside the motor;

[0091] The maximum temperature data of the motor surface, the stator winding temperature data, the rotor core temperature data, the bearing temperature data, and the cooling medium temperature data are weighted and averaged to obtain the comprehensive temperature data. The weight of each data is 0.2. Each temperature data contributes the same to the comprehensive temperature and is considered equally important.

[0092] Color mapping is a method of converting the RGB value of each pixel in a thermal image into corresponding temperature data. By mapping the pixel value of the thermal image to a color-temperature table, the specific temperature of each pixel can be obtained.

[0093] The motor surface temperature and multi-point temperature information of key internal components are combined to comprehensively reflect the overall thermal status of the motor through weighted average.

[0094] The methods for obtaining preprocessed datasets include:

[0095] Based on the acquired motor mechanical performance data and motor electrical performance data, the interpolation method is used to handle missing values, and the Z-Score method is used to handle outliers to obtain a preliminary data set;

[0096] Based on the preliminary data set, the dynamic error constrained recursive filtering method is used to remove noise and obtain the filter output data set;

[0097] The filter output data set is then normalized and the timestamps are unified to obtain the preprocessed data set.

[0098] The specific method of using the dynamic error constrained recursive filtering method to remove noise and obtain the filter output data set includes:

[0099] Step o1: take the preliminary dataset as input;

[0100] Step o2: Initialize the parameters of the dynamic error constrained recursive filtering method, including filter coefficients, forgetting factors, and error covariance matrix;

[0101] Step o3: Calculate the gain data k(n) at the nth moment. The formula is: P(n-1) represents the covariance matrix at time n-1, x(n) represents the preliminary data set at time n, and x(n) T represents the transpose of the preliminary data set at time n, and λ represents the forgetting factor;

[0102] Update the error covariance matrix, the formula is: Where P(n) represents the updated error covariance matrix;

[0103] The filter output data set is calculated using the filter coefficients and the preliminary data set, using the formula: s(n) = w(n-1) T x(n), where w(n-1) T represents the transpose of the filter coefficient at time n-1, and s(n) represents the filter output data set at time n;

[0104] The error data is calculated by the difference between the expected data set and the filter output data set. The formula is: e(n) = y(n) - s(n), where e(n) represents the error data at time n, and y(n) represents the expected data set including the reference data without noise, which is obtained by consulting relevant materials.

[0105] Update the filter coefficients based on the gain data and error data. The formula is: w(n) = w(n-1) + k(n)e(n);

[0106] Step o4: Repeat the above update steps, set the error threshold to m_m, and stop the iteration when the error data is less than the set error threshold to m_m, and obtain the filter output data set, where the error threshold to m_m is obtained by taking the average value of K_O experiments;

[0107] The preliminary data set includes vibration data, speed data, current data, comprehensive temperature data, voltage data, and power data;

[0108] The expected data set includes vibration data without noise, speed data, current data, integrated temperature data, voltage data, and power data;

[0109] The filter output data set is the vibration data, speed data, current data, integrated temperature data, voltage data and power data after noise removal by the dynamic error constrained recursive filtering method;

[0110] The dynamic error-constrained recursive filtering method is used to effectively suppress the noise in the preliminary data set and improve the quality and accuracy of the data. The motor mechanical and electrical performance data are usually interfered with by various external factors, resulting in noise in the data affecting the prediction and analysis results. The dynamic error-constrained recursive filtering method can adjust and remove this noise in real time by dynamically updating the filter coefficients, extracting more accurate data and providing a more reliable foundation for subsequent data predictions.

[0111] Specific methods for using multimodal drive enhancement regression technology to predict motor load data include:

[0112] Step M1: The input sample set includes h_h samples, and the input sample set D={X i ,Y i}, each sample includes a preprocessed data set and corresponding motor load data;

[0113] where X i represents the i-th sample preprocessing data set, Y i represents the i-th sample motor load data, the data in the preprocessed data set is the feature variable, and the motor load data is the target variable;

[0114] Step M2: Set the maximum number of iterations Td and initialize the sample weights

[0115] Step M3, iterative training of weak learners: In the tth round of iteration, use the current sample weight To train the weak regressor, the weak learner is a linear regression model, the formula is: in, represents the weight of the i-th sample in the t-th round, b is the bias term of the weak learner, h t (X i ) is the motor load prediction data predicted by the weak learner for the i-th sample in the t-th iteration, and t represents the index of the iteration number;

[0116] Calculate the weighted error of the weak learner, the formula is: Among them, ∈ t represents the weighted error in the tth iteration;

[0117] The weighted error of the weak learner is used to calculate the weight of the weak learner. The formula is: Among them, α t Represents the weight of the weak learner in the tth iteration. When the error is small, αt A larger value of indicates that the regressor contributes more to the final model;

[0118] Based on the weighted error of the weak learner, the weight of each sample is updated as follows: represents the weight of the i-th sample in the t-th round, represents the updated weight of the i-th sample in the t-th round;

[0119] Step M4: Repeat step M3 until the maximum number of iterations Td is reached and then stop iterating;

[0120] Step M5: construct a strong learner by weighting the motor load prediction data predicted by the weak learner. The formula is: G(X i ) is a strong learner, representing the final output motor load data;

[0121] Each round of iteration obtains motor load prediction data predicted by a weak learner, and the number of weak learners is consistent with the number of iterations.

[0122] The specific methods of constructing the spatiotemporal correlation graph include:

[0123] The node system of the spatiotemporal correlation graph consists of physical entity nodes and virtual causal nodes. The physical entity nodes include motor load data and motor environment data. The virtual causal nodes include load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data.

[0124] Among them, load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data are calculated by Granger causality analysis method;

[0125] The nodes in the acquired node system are timestamped uniformly, the node system with the same timestamp is extracted, and each data in the node system is normalized; the weights between the object entity nodes and the virtual causal nodes in the node system are calculated using the ST-CPF method;

[0126] The specific method of ST-CPF is as follows:

[0127] Use matplotlib to draw a scatter plot to determine whether the relationship between nodes is linear or nonlinear; if it is linear, use the Pearson correlation method to calculate the correlation between nodes; if it is nonlinear, use the mutual information method to calculate the correlation between nodes; generate a correlation matrix;

[0128] Use Euclidean distance to calculate the distance between nodes; generate a distance matrix;

[0129] Based on the correlation matrix and its proximity matrix, the weights between nodes are calculated using the linear weighting method;

[0130] A spatiotemporal correlation graph is formed based on the nodes, the weights between nodes, and the corresponding timestamps.

[0131] The specific methods of embedding the spatiotemporal correlation graph into the multi-criteria Gaussian detection method to calculate the comprehensive probability density include:

[0132] Step V1: Input the spatiotemporal correlation graph and treat the physical entity node and virtual causal node corresponding to the same timestamp in the spatiotemporal correlation graph as a data point;

[0133] Step V2, initialize the mean, covariance matrix and mixing weight parameters of the Gaussian component;

[0134] Step V3, using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components;

[0135] Step V4: Calculate the probability density of each Gaussian component, and use the probability density of the Gaussian component to calculate the responsibility of the Gaussian component;

[0136] Probability density: Wherein, d is the dimension of the covariance matrix, that is, the number of motor load prediction data and motor environment data; the motor environment data includes corrosive gas concentration data, electromagnetic intensity data, ambient temperature data, ambient humidity data and ambient noise data, 5 data, so the dimension of the covariance matrix is ​​d=6; μ H is the mean of the Hth Gaussian component, H is the index of the Gaussian component, β H is the covariance matrix of the Hth Gaussian component, |β H | is the determinant of the covariance matrix, is the inverse matrix of the covariance matrix, x_x c is the cth data point, q(x_x c |μ H ,β H ) is the probability density of the Hth Gaussian component of the cth data point;

[0137] The responsibility formula is: Where F is the total number of Gaussian components, γ c,H is the responsibility of the Hth Gaussian component of the cth data point; π H is the initialization mixing weight of the Hth Gaussian component;

[0138] Step V5: Update the mean, covariance matrix and mixing weight parameters of the Gaussian component according to the degree of responsibility; the updated mean is: Where N is the total number of data points; update the covariance matrix: Update the blending weights:

[0139] Step V6: Repeat steps V4 and V5 until the set number of iterations is reached and the update stops;

[0140] Step V7: Calculate the comprehensive probability density of each data point using the formula: Among them, Q(x_x c ) is the comprehensive probability density of the c-th data point, is the updated probability density of the Hth Gaussian component.

[0141] The specific methods of using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components include:

[0142] In the multi-criteria Gaussian detection method, the optimal number of Gaussian components is selected by AIC and BIC to avoid overfitting, that is, selecting too many Gaussian components; or underfitting, that is, selecting too few Gaussian components;

[0143] The candidate value range of the number of Gaussian components is set according to the empirical rule. For each candidate number of Gaussian components, the log-likelihood value of the Gaussian mixture model is calculated. The formula is: Among them, S is the number of candidate Gaussian components, s_b is the index of the number of candidate Gaussian components, π s_b represents the mixing weight of the s_bth Gaussian component, μ s_b represents the mean of the s_bth Gaussian component, β s_b represents the covariance matrix of the s_bth Gaussian component, q(x_x c |μ s_b ,β s_b ) is the probability density of the s_bth Gaussian component of the cth data, and lnL is the log-likelihood value of the number of candidate Gaussian components S;

[0144] Using the log-likelihood value, calculate the AIC value, AIC(S) = 2*s_L-2*lnL, where s_L is the number of parameters of the Gaussian mixture model; calculate the BIC value, BIC(S) = s_L*lnN-2*lnL,

[0145] The calculation formula for the number of parameters of the multi-criteria Gaussian detection method is: Where d is the number of mean parameters, is the number of covariance matrix parameters, 1 is the number of mixing weight parameters;

[0146] The weighted average of the AIC and BIC values ​​of the number of candidate Gaussian components S is used to obtain the comprehensive index: C_o(S)=w AIC *AIC(S)+w BIC *BIC(S), where w AIC and w BIC are the weight coefficients of AIC and BIC respectively;

[0147] By comparing the comprehensive indices of different candidate Gaussian component numbers, the candidate value with the smallest comprehensive index is selected as the optimal Gaussian component number.

[0148] The specific methods of setting a standard threshold and detecting whether the motor operation is abnormal by comparing the comprehensive probability density with the standard threshold include:

[0149] Set the initial threshold by using the historical comprehensive probability density: G_e=J_Z Q +Yz*B_C Q , where G_e is the initial threshold, J_Z Q is the average value of the historical comprehensive probability density, B_C Q is the standard deviation of the historical comprehensive probability density, and Yz is the adjustment coefficient, which is obtained by empirical method;

[0150] Calculate the historical reconstruction data points by the responsibility and mean of the Gaussian components: in, is the reconstructed value of the cth historical data point;

[0151] The reconstruction error between historical data points and historical reconstructed data points is calculated using the Euclidean distance. The formula is: Among them, E c represents the reconstruction error of the c-th historical data point, x_x c,wd Represents the value of the cth historical data point in the wdth dimension, Represents the value of the cth historical reconstruction data point in the wdth dimension;

[0152] The initial threshold is modified by calculating the average value of the reconstruction error to obtain the standard threshold: Among them, B_Y is the standard threshold, is the average value of the reconstruction error;

[0153] The calculation formula for the average reconstruction error is:

[0154] The purpose of using the average value of the reconstruction error to correct the initial threshold is to combine the degree of data deviation and make the threshold more suitable for the actual situation. This can more accurately distinguish normal and abnormal data, reduce misjudgment, and improve detection reliability.

[0155] When a new set of data points is input, a new comprehensive probability density is obtained through the Gaussian mixture model. When the new comprehensive probability density is greater than or equal to the standard threshold, the motor is operating normally; when the new comprehensive probability density is less than the standard threshold, the motor is operating abnormally and an early warning is issued.

[0156] Assuming that the comprehensive probability density of the new data point is greater than or equal to the standard threshold, it means that it is consistent with the distribution of historical data and the motor is operating normally; if it is lower than the threshold, it means that the data point deviates from the normal distribution, an abnormality occurs, and the motor needs an early warning; this method is based on the principle of probability density in statistics: data in normal state usually has a higher probability density, and data in abnormal state will produce a lower probability density. Therefore, by comparing with the distribution of historical data, it is possible to effectively determine whether the motor is in an abnormal state.

[0157] This embodiment can comprehensively combine the motor's mechanical performance, electrical performance, and motor environmental data to perform comprehensive preprocessing, aiming to improve the motor's operating stability. A dynamic error-constrained recursive filtering method is used to remove noise, thereby ensuring the accuracy of the processed data and laying a solid foundation for subsequent motor load prediction and operating status diagnosis.

[0158] Multimodal drive enhanced regression technology is used to predict motor load data. This method can effectively improve the accuracy of the prediction model because each round of weak learners optimizes the error of the previous round, gradually reducing the model's deviation and providing more accurate load predictions, thereby effectively assisting in motor operating status assessment and early warning.

[0159] In terms of anomaly detection, a multi-criteria Gaussian detection method is adopted. This method determines whether the motor has an abnormality by calculating the comprehensive probability density of motor load data and motor environment data. The AIC and BIC criteria are used to optimize the number of Gaussian components to ensure the rationality and accuracy of the model structure. In addition, the initial threshold is adjusted by calculating the reconstruction error between the data point and the historical reconstructed data point to obtain the standard threshold, which makes the motor anomaly detection more sensitive and robust.

[0160] Example 2, please refer to Figure 2 As shown, for the parts not described in detail in this embodiment, please refer to the description of embodiment 1. An industrial automation motor detection system is provided, including:

[0161] Multi-source data acquisition module: acquires motor mechanical performance data, motor electrical performance data, and motor environmental data; it generates comprehensive temperature data by fusing infrared thermal imaging with embedded sensors;

[0162] Dynamic filtering processing module: This module uses a dynamic error-constrained recursive filtering method to denoise the motor mechanical and electrical performance data, obtain a filter output dataset, and generate a preprocessed dataset with a unified time and space reference. The dynamic error-constrained recursive filtering method obtains the filter output dataset by continuously updating the filter coefficients and error covariance matrix.

[0163] Motor Load Prediction Module: A multimodal drive-enhanced regression technique is constructed for preprocessed datasets. Feature selection is performed on the preprocessed dataset by iteratively optimizing the sample weight distribution, and motor load data is predicted. This technique optimizes sample weights by iteratively training weak learners and weighting sample errors. Finally, a strong learner is constructed through a weighted combination of multiple weak learners to predict motor load data.

[0164] Spatiotemporal correlation multi-criteria early warning module: Based on the spatiotemporal correlation of motor load data and motor environment data, a spatiotemporal correlation graph is constructed; the spatiotemporal correlation graph is embedded in the multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points; and a standard threshold is set. When the probability density of real-time data points deviates from the standard threshold, a graded early warning mechanism is triggered; in the multi-criteria Gaussian detection method, the Akaike Information Criterion (AIC) and the Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

[0165] Example 3. This embodiment discloses an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the operation mode of the industrial automation motor detection method and system provided above is realized.

[0166] Since the electronic device described in this embodiment is an electronic device used to implement an industrial automation motor detection method and system in the embodiments of this application, based on the industrial automation motor detection method and system described in the embodiments of this application, those skilled in the art will be able to understand the specific implementation of the electronic device of this embodiment and its various variations. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. As long as those skilled in the art implement the electronic device used in the industrial automation motor detection method and system in the embodiments of this application, they are within the scope of protection to be provided by this application.

[0167] The above formulas are all dimensionless and numerical calculations. The formulas are obtained by collecting a large amount of data and performing software simulation to obtain the most recent real situation. The preset parameters and thresholds in the formulas are set by technicians in this field according to actual conditions.

[0168] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiment. All technical solutions based on the principles of the present invention are within the scope of protection of the present invention. It should be noted that for users of ordinary skill in the art, various improvements and modifications that do not depart from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An industrial automation motor detection method, characterized in that: include: Step A1: Acquire motor mechanical performance data, motor electrical performance data, and motor environmental data; Among them, comprehensive temperature data is generated by fusion of infrared thermal imaging and embedded sensors; Step A2: De-noising the motor mechanical performance data and the motor electrical performance data using a dynamic error constrained recursive filtering method to obtain a filter output data set and generate a pre-processed data set; The dynamic error constrained recursive filtering method obtains the filter output data set by continuously updating the filter coefficients and the error covariance matrix; Step A3: construct a multimodal drive enhanced regression technology for the preprocessed dataset, perform feature selection on the preprocessed dataset by iteratively optimizing the sample weight distribution, and predict the motor load data; Multimodal drive enhanced regression technology optimizes sample weights by iteratively training weak learners and weighting sample errors. Ultimately, a strong learner is constructed through a weighted combination of multiple weak learners to predict motor load data. Step A4: performing spatiotemporal correlation based on the motor load data and the motor environment data to construct a spatiotemporal correlation graph; embedding the spatiotemporal correlation graph into a multi-criteria Gaussian detection method to calculate the comprehensive probability density of the data points; And set a standard threshold, when the probability density of real-time data points deviates from the standard threshold, trigger the graded warning mechanism; In the multi-criteria Gaussian detection method, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

2. The industrial automation motor detection method according to claim 1, characterized in that: The motor mechanical performance data includes vibration data and speed data; Motor electrical performance data includes current data, comprehensive temperature data, voltage data and power data; The motor environment data includes corrosive gas concentration data, electromagnetic intensity data, ambient temperature data, ambient humidity data, and ambient noise data.

3. The industrial automation motor detection method according to claim 2, characterized in that: The method for obtaining the comprehensive temperature data includes: Use an infrared thermal imager to obtain a thermal image of the motor surface. Use OpenCV to process the thermal image. Extract the RGB value of each pixel from the thermal image and convert it into temperature data through color mapping. Store the temperature data of each pixel in a temperature matrix. Get the maximum temperature value in the temperature matrix and define it as the maximum temperature data of the motor surface. Use embedded temperature sensors to collect stator winding temperature data, rotor core temperature data, bearing temperature data, and cooling medium temperature data inside the motor; The maximum temperature data of the motor surface, the stator winding temperature data, the rotor core temperature data, the bearing temperature data and the cooling medium temperature data are weighted averaged to obtain the comprehensive temperature data.

4. The industrial automation motor detection method according to claim 3, characterized in that: The method of obtaining the preprocessed data set includes: Based on the acquired motor mechanical performance data and motor electrical performance data, the interpolation method is used to handle missing values, and the Z-Score method is used to handle outliers to obtain a preliminary data set; Based on the preliminary data set, the dynamic error constrained recursive filtering method is used to remove noise and obtain the filter output data set; The filter output data set is then normalized and the timestamps are unified to obtain the preprocessed data set.

5. The industrial automation motor detection method according to claim 4, characterized in that: The specific method of using the dynamic error constrained recursive filtering method to remove noise and obtain the filter output data set includes: Step o1: take the preliminary dataset as input; Step o2: Initialize the parameters of the dynamic error constrained recursive filtering method, including filter coefficients, forgetting factors, and error covariance matrix; Step o3, calculating the gain data at the nth moment by using the covariance weighted gain solution method based on the preliminary data set at the nth moment, the transpose of the preliminary data set at the nth moment, the error covariance matrix at the n-1th moment, and the forgetting factor; Update the error covariance matrix based on the gain data and the error covariance matrix at time n-1; calculating a filter output data set using the filter coefficients and the preliminary data set; Calculate error data using the expected data set and the filter output data set; updating filter coefficients according to the gain data and the error data; Step o4: Repeat the above updating steps, set the error threshold to m_m, and when the error data is less than the set error threshold to m_m, stop the iteration and obtain the filter output data set.

6. The industrial automation motor detection method according to claim 5, characterized in that: The specific method of using the multimodal drive enhanced regression technology to predict motor load data includes: Step M1: The input sample set includes h_h samples, and each sample in the input sample set includes a preprocessed data set and corresponding motor load data; the data in the preprocessed data set is the feature variable, and the motor load data is the target variable; Step M2: Set the maximum number of iterations and initialize the sample weights; Step M3, iteratively training the weak learner: in the tth iteration, the current sample weight is used to train the weak regressor to obtain the motor load prediction data predicted by the weak learner; Based on the difference between the motor load prediction data and the motor load data predicted by the weak learner for the sample in the t-th iteration, combined with the corresponding sample weights, the weighted error of the weak learner in the current round is obtained by the weighted error method; The weighted error of the weak learner is used to calculate the weight of the weak learner; Based on the weighted error of the weak learner, update the weight of each sample; Step M4: Repeat step M3 until the maximum number of iterations is reached and stop the iteration; Step M5: constructing a strong learner by weighting the motor load prediction data predicted by the weak learner, that is, the motor load data finally outputted.

7. The industrial automation motor detection method according to claim 6, characterized in that: The specific method of constructing the spatiotemporal association graph includes: The node system of the spatiotemporal correlation graph consists of physical entity nodes and virtual causal nodes. The physical entity nodes include motor load data and motor environment data. The virtual causal nodes include load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data. Among them, load-gas data, load-intensity data, load-temperature data, load-humidity data, and load-noise data are calculated by Granger causality analysis method; The nodes in the acquired node system are timestamped uniformly, the node system with the same timestamp is extracted, and each data in the node system is normalized; the weights between the object entity nodes and the virtual causal nodes in the node system are calculated using the ST-CPF method; The specific method of ST-CPF is as follows: Use matplotlib to draw a scatter plot to determine whether the relationship between nodes is linear or nonlinear; if it is linear, use the Pearson correlation method to calculate the correlation between nodes; if it is nonlinear, use the mutual information method to calculate the correlation between nodes; generate a correlation matrix; Use Euclidean distance to calculate the distance between nodes; generate a distance matrix; Based on the correlation matrix and its proximity matrix, the weights between nodes are calculated using the linear weighting method; A spatiotemporal correlation graph is formed based on the nodes, the weights between nodes, and the corresponding timestamps.

8. The industrial automation motor detection method according to claim 7, characterized in that: The specific method of embedding the spatiotemporal correlation graph into the multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points includes: Step V1: Input the spatiotemporal correlation graph and treat the physical entity node and virtual causal node corresponding to the same timestamp in the spatiotemporal correlation graph as a data point; Step V2, initialize the mean, covariance matrix and mixing weight parameters of the Gaussian component; Step V3, using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components; Step V4: Calculate the probability density of each Gaussian component, and use the probability density of the Gaussian component to calculate the responsibility of the Gaussian component; Step V5: updating the mean, covariance matrix and mixing weight parameters of the Gaussian component according to the degree of responsibility; Step V6: Repeat steps V4 and V5 until the set number of iterations is reached and the update stops; Step V7: Calculate the comprehensive probability density of each data point.

9. The industrial automation motor detection method according to claim 8, characterized in that: The specific method of using Akaike Information Criterion AIC and Bayesian Information Criterion BIC to define the optimal number of Gaussian components includes: According to the empirical rule, the candidate value range of the number of Gaussian components is set. For each candidate number of Gaussian components, the log-likelihood value of the multi-criteria Gaussian detection method is calculated. The AIC value and BIC value are calculated using the log-likelihood value. The AIC value and BIC value are weighted averaged to obtain a comprehensive index. By comparing the comprehensive indexes of different candidate numbers of Gaussian components, the candidate value with the smallest comprehensive index is selected as the optimal number of Gaussian components. For each data point, the probability density of each candidate Gaussian component of each data point is calculated; then, the probability density of the comprehensive candidate Gaussian component of each data point is calculated by weighted averaging based on the mixing weight corresponding to the probability density of each Gaussian component; based on the natural logarithm method, the probability densities of the comprehensive candidate Gaussian components of N data points are added to obtain the log-likelihood value; The AIC value is calculated based on twice the number of parameters of the Gaussian mixture model and twice the log-likelihood value; The BIC value is calculated by multiplying the number of parameters and the log-likelihood value of the multi-criteria Gaussian detection method by the overall log-likelihood value times 2; Among them, the number of parameters of the multi-criteria Gaussian detection method is obtained by multiplying the number of mean parameters, the number of mixing weight parameters, and the number of covariance matrix parameters constructed by the number of mean parameters by the number of candidate Gaussian components.

10. The industrial automation motor detection method according to claim 9, characterized in that: The specific method of setting the standard threshold and detecting whether the operation of the motor is abnormal by comparing the comprehensive probability density with the standard threshold includes: The initial threshold is set by the historical comprehensive probability density; the responsibility and mean of the Gaussian component are used to calculate the historical reconstructed data points; the reconstruction error between the historical data points and the historical reconstructed data points is obtained based on the Euclidean distance; and the average value of the reconstruction error is used to correct the initial threshold to obtain the standard threshold. A new set of data points is input and a new comprehensive probability density is obtained through the Gaussian mixture model. When the new comprehensive probability density is greater than or equal to the standard threshold, the motor operates normally; when the new comprehensive probability density is less than the standard threshold, the motor operates abnormally and an early warning is issued.

11. An industrial automation motor detection system, based on an industrial automation motor detection method according to any one of claims 1 to 10, characterized in that: include: Multi-source data acquisition module: acquires motor mechanical performance data, motor electrical performance data, and motor environmental data; Among them, comprehensive temperature data is generated by fusion of infrared thermal imaging and embedded sensors; Dynamic filtering processing module: It uses dynamic error constraint recursive filtering method to denoise the motor mechanical performance data and motor electrical performance data, obtains the filter output data set, and generates a preprocessing data set with a unified time and space reference; The dynamic error constrained recursive filtering method obtains the filter output data set by continuously updating the filter coefficients and the error covariance matrix; Motor load prediction module: Based on the preprocessed dataset, a multimodal drive enhancement regression technology is constructed to perform feature selection on the preprocessed dataset by iteratively optimizing the sample weight distribution, and predict the motor load data; Multimodal drive enhanced regression technology optimizes sample weights by iteratively training weak learners and weighting sample errors. Ultimately, a strong learner is constructed through a weighted combination of multiple weak learners to predict motor load data. Spatiotemporal correlation multi-criteria early warning module: Based on the spatiotemporal correlation of motor load data and motor environment data, a spatiotemporal correlation graph is constructed. The spatiotemporal correlation graph is embedded in the multi-criteria Gaussian detection method to calculate the comprehensive probability density of data points. And set a standard threshold, when the probability density of real-time data points deviates from the standard threshold, trigger the graded warning mechanism; In the multi-criteria Gaussian detection method, Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) are used to define the optimal number of Gaussian components.

Images