Current fluctuation anomaly detection method based on machine learning LSTM (Long Short Term Memory) and SVM (Support Vector Machine)

Abstract

The invention provides a current fluctuation anomaly detection method based on machine learning LSTM (Long Short Term Memory) and SVM (Support Vector Machine). The method comprises a data acquisition and preprocessing step, a feature extraction step, a machine learning model training step, an anomaly detection and alarm step and a remote monitoring step. Wherein the machine learning model training step comprises SVM training and LSTM training; the anomaly detection and alarm comprises the following steps: SVM real-time classification: inputting a signal which is acquired and preprocessed in real time into a trained SVM model, and outputting an alarm signal if the SVM model judges that the running state of the motor is abnormal; and LSTM trend prediction: inputting the calculated time sequence data into the trained LSTM model, and if the LSTM model predicts that the future current has abnormal fluctuation, sending out an early warning signal in advance. According to the invention, real-time monitoring and anomaly detection of the running state of the motor can be realized, the safety and stability of equipment are improved, the risk of fault shutdown is reduced, and the production efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the fields of industrial automation and electric vehicle technology, and specifically to a current fluctuation anomaly detection method based on machine learning LSTM and SVM, which is suitable for real-time monitoring and fault warning of motors in the fields of industrial automation, electric vehicles, robots, etc. Background Art

[0002] As the core power unit in fields like industrial automation and electric vehicles, the operating status of motors plays a decisive role in the safety, stability, and energy consumption of these devices. In industrial production, various types of mechanical equipment driven by motors are widely used on production lines, and the stable operation of motors is key to ensuring production efficiency and product quality. In the electric vehicle sector, motor performance directly affects the vehicle's range, power output, and driving safety. Therefore, real-time monitoring of the motor's operating status and accurate detection of abnormal load changes (such as overload, bearing wear, and jamming) are of paramount practical significance, effectively improving equipment life and production efficiency.

[0003] Currently, the motor load abnormality detection methods mainly include the following:

[0004] Fixed threshold method: This method determines whether the motor has an abnormality by setting fixed upper and lower limits for current changes. However, this method has obvious limitations. On the one hand, it is only applicable to specific working conditions. When the motor load is in a dynamic change, the fixed threshold cannot adapt to such changes, which can easily lead to false alarms or missed alarms. For example, during the motor startup phase or when the load suddenly increases, the current may briefly exceed the fixed threshold, but in fact the motor has not failed, which will result in a false alarm; and in some slowly developing faults, because the current change does not exceed the fixed threshold, missed alarms will occur. On the other hand, fixed thresholds make it difficult to detect trend faults and cannot detect the gradual deterioration of motor performance in advance.

[0005] Frequency domain analysis method (Fourier transform FFT): The frequency domain analysis method mainly identifies anomalies of specific frequencies, such as specific frequency harmonics caused by bearing wear, by analyzing the spectral characteristics of the current signal. However, this method also has certain limitations. It is mainly suitable for the detection of periodic faults and can play a good role in faults with periodic characteristics such as bearing wear. However, for non-periodic faults, such as jamming and instantaneous overload, the frequency domain analysis method is difficult to detect effectively. In addition, frequency domain analysis requires a longer acquisition time window to obtain accurate spectral information, which results in its poor real-time performance and cannot meet the needs of real-time monitoring of the motor operating status.

[0006] Fault diagnosis based on physical modeling: This method establishes a mathematical model of the motor, calculates theoretical values ​​for parameters such as load, current, and voltage, and compares these theoretical values ​​with the actual values ​​to determine whether the motor is faulty. However, this method has some shortcomings: First, it requires a precise motor model and is only applicable to specific motor types, resulting in poor generalization. Different types of motors have different characteristics and parameters, making it difficult to establish a universal physical model. Second, under complex operating conditions, the actual operating state of the motor can be affected by a variety of factors, resulting in reduced model accuracy and a high false alarm rate.

[0007] With the rapid development of artificial intelligence (AI) and machine learning (ML), an increasing number of intelligent detection methods are being applied to motor load anomaly detection. Support vector machines (SVM) and long short-term memory (LSTM) networks have become relatively advanced detection methods. SVM uses nonlinear classification to classify current signal features and detect anomalies. LSTM, a deep learning model optimized for time series data, is suitable for motor signal trend prediction and anomaly detection.

[0008] However, the existing motor load abnormality detection methods still have the following technical problems:

[0009] High false alarm rate and difficulty adapting to dynamic load changes: The traditional fixed threshold method cannot adapt to the dynamic load changes of the motor under different working conditions, resulting in frequent false alarms or missed alarms, affecting the accuracy of detection.

[0010] Unable to predict failures in advance and lack of preventive maintenance capabilities: Traditional detection methods are mostly based on the instantaneous value of the current signal. They can only detect failures that have already occurred, but cannot predict possible future failures of the equipment in advance. Preventive maintenance cannot be implemented, increasing the risk of equipment failure and downtime.

[0011] It is difficult to comprehensively analyze multiple abnormal modes based on time domain or frequency domain analysis only: Existing methods usually perform analysis based on a single feature (time domain or frequency domain). For example, FFT can only detect periodic faults, and the fixed threshold method is only applicable to static loads. It is impossible to comprehensively and integratedly analyze the various abnormal modes of the motor.

[0012] Traditional methods are computationally intensive and difficult to meet real-time monitoring needs: physical methods based on mathematical modeling or complex FFT analysis require a large amount of computing resources, resulting in slow detection speeds, making it difficult to achieve real-time monitoring of industrial sites and unable to meet the real-time requirements of industrial production.

[0013] In summary, the existing motor load anomaly detection methods have many shortcomings in terms of accuracy, real-time performance, comprehensive analysis capabilities, and preventive maintenance. There is an urgent need for a more efficient, accurate, and comprehensive motor operation status monitoring and anomaly detection method. Summary of the Invention

[0014] In response to the above-mentioned technical problems in the prior art, the present invention provides a current fluctuation anomaly detection method based on machine learning LSTM and SVM. This method can realize real-time monitoring and anomaly detection of the motor operating status, improve the safety and stability of the equipment, reduce the risk of fault downtime, and improve production efficiency.

[0015] The present invention achieves the above-mentioned purpose through the following technical solutions:

[0016] A method for detecting abnormal current fluctuations based on machine learning LSTM and SVM, comprising the following steps:

[0017] Data acquisition and preprocessing: Collect the motor's current signal and load signal, and preprocess the collected signals;

[0018] Feature extraction: Calculate time domain features to determine load offset, detect abnormal loads, identify abnormal load states, and detect sudden faults; perform frequency domain feature extraction, using fast Fourier transform to obtain the signal's spectral characteristics to identify abnormal motor frequencies and the degree of nonlinear distortion; calculate time series features, using short-time Fourier transform and wavelet transform to analyze the signal, obtain the frequency characteristics of the signal that change over time, and detect short-time impulse signals or sudden faults;

[0019] Machine learning model training: This includes support vector machine (SVM) training and long-term time-lapse (LSTM) training. The trained SVM model is used to determine the motor's operating status in real time, and the trained LSTM model is used to predict future current trends.

[0020] Abnormal detection and alarm:

[0021] SVM real-time classification: The real-time collected and pre-processed signals are input into the trained SVM model. If the SVM model determines that the motor operation status is abnormal, an alarm signal is output;

[0022] LSTM trend prediction: The calculated time series data is input into a trained LSTM model. If the LSTM model predicts abnormal current fluctuations in the future, an early warning signal is issued.

[0023] According to a method for detecting abnormal current fluctuations based on machine learning LSTM and SVM, the preprocessing includes:

[0024] The collected signals are subjected to noise reduction processing, low-pass filtering is used to remove high-frequency noise, and mean filtering is used to smooth the signals; the data are normalized to map the data to a specific range; and a sliding window method is used to construct a time series data set;

[0025] in, I norm is the normalized value, I is the acquired current value, I min is the minimum current, I max is the maximum current value, so that the data is mapped to [0,1].

[0026] According to a current fluctuation anomaly detection method based on machine learning LSTM and SVM provided by the present invention, the time domain features include:

[0027] Mean: Indicates the central trend of the signal and determines whether the load is offset. N is the number of sampling points, I i is the current value of the i-th sampling point;

[0028] Standard Deviation: Indicates the degree of signal fluctuation and detects abnormal loads.

[0029] Skewness: measures the symmetry of the signal distribution and identifies abnormal load conditions. If Skewness>0, it indicates that the distribution is right-skewed, and if Skewness<0, it indicates that the distribution is left-skewed;

[0030] Kurtosis: measures whether the signal has prominent peaks and can be used to detect sudden faults. Kurtosis>3 indicates abnormal spikes, and Kurtosis<3 indicates a relatively flat signal.

[0031] According to a method for detecting abnormal current fluctuations based on machine learning LSTM and SVM provided by the present invention, the spectral characteristics of the signal are obtained by fast Fourier transform, including:

[0032] FFT transformation: extract the spectrum components of the current signal and identify abnormal motor frequencies. Among them F (k) is the amplitude of the kth frequency component, I(n) is the time domain signal, and N is the number of sampling points;

[0033] Harmonic analysis: Total harmonic distortion is used to measure the degree of nonlinear distortion of the current signal. F (1) is the fundamental wave, F (k) is the kth harmonic component.

[0034] According to the present invention, a current fluctuation anomaly detection method based on machine learning LSTM and SVM is provided, which uses short-time Fourier transform (STFT) and wavelet transform (WT) to analyze the signal, including:

[0035] Short-time Fourier transform STFT: analyzes the frequency characteristics of signals changing over time, Where ω(mn) is the window function, SIFT (n,k) is the k frequency components of the nth time segment;

[0036] Wavelet transform WT: detect short-time impulse signals or sudden faults, in is the mother wavelet function, a is the control scale, and b is the control translation.

[0037] According to the present invention, a current fluctuation anomaly detection method based on machine learning LSTM and SVM is provided, wherein the SVM training and LSTM training include:

[0038] SVM model training: The motor operating status is labeled as normal or abnormal based on historical data. The labeled data is used to train the support vector machine (SVM). The Gaussian kernel function is used for classification, and the SVM parameters are optimized through cross-validation.

[0039] LSTM model training: The LSTM model structure is designed based on the constructed time series features, and the mean square error (MSE) is used as the loss function to train the LSTM model.

[0040] According to the present invention, a method for detecting abnormal current fluctuations based on machine learning LSTM and SVM is provided, wherein the LSTM model structure includes:

[0041] Input layer: receives T historical current data as input, where T is the preset number of historical data, and the historical current data is pre-processed motor current signal data;

[0042] Hidden layer: includes h layers of LSTM units. Each layer of LSTM units includes g neurons, where h is an integer greater than or equal to 1 and g is a positive integer. Each layer of LSTM units is connected in sequence to perform deep feature extraction and memory of the input historical current data.

[0043] Output layer: outputs the predicted current value at 1 to 5 time points in the future. The output layer calculates the predicted current value at different time points in the future through the fully connected layer based on the features extracted by the hidden layer.

[0044] According to the present invention, a method for detecting abnormal current fluctuations based on machine learning LSTM and SVM also includes:

[0045] Using the MQTT / HTTP protocol, the collected current signals, load signals, SVM classification results, and LSTM prediction results are uploaded to the cloud IoT platform for remote data analysis, including:

[0046] The uploaded data is stored and a historical database is established. Big data analysis algorithms are used to conduct in-depth mining of the stored data and analyze the correlation between current fluctuations and load changes. Visual reports are generated based on the analysis results to display trends and abnormalities in the motor's operating status.

[0047] According to a current fluctuation anomaly detection method based on machine learning LSTM and SVM provided by the present invention, when uploading data to the cloud IoT platform, the transmitted data is encrypted and a hybrid encryption method is adopted:

[0048] First, the data is encrypted using the Advanced Encryption Standard AES algorithm;

[0049] The AES encryption key is then encrypted using the RSA algorithm;

[0050] Upload the encrypted data and the encrypted AES key to the cloud IoT platform;

[0051] On the cloud IoT platform, the RSA private key is first used to decrypt the AES encryption key, and then the decrypted AES encryption key is used to decrypt the data for remote data analysis.

[0052] According to a current fluctuation anomaly detection method based on machine learning LSTM and SVM provided by the present invention, when using the SVM model, the calculated time domain features, frequency domain features and time series features are combined into an input feature vector, and the input feature vector is input into a trained support vector machine SVM model. The SVM model uses a Gaussian kernel function, and the SVM model performs classification and judgment based on normal and abnormal labels pre-annotated based on historical data. If the output result is an abnormal category, the motor operation state is determined to be abnormal.

[0053] Therefore, the present invention proposes a current fluctuation anomaly detection method based on machine learning LSTM and SVM, which has significant beneficial effects compared with traditional motor anomaly detection technology, specifically reflected in the following aspects:

[0054] 1. Significantly improve the accuracy of motor anomaly detection and significantly reduce false positives and missed alarms: This invention uses a machine learning model for adaptive learning, automatically adjusting the detection threshold based on the actual operating data of the motor under different operating conditions, effectively improving adaptability to complex operating conditions. By combining time domain, frequency domain, and time-frequency domain feature analysis, this invention can comprehensively and accurately detect various types of motor load anomalies, including but not limited to overload, bearing wear, and jamming, thereby significantly reducing false positives and missed alarms.

[0055] 2. Powerful Fault Trend Prediction Capabilities, Enabling Preventive Maintenance: Leveraging the time series modeling capabilities of LSTM (Long Short-Term Memory) networks, this invention can deeply analyze historical current data and predict future trends, identifying potential motor failures in advance, thus shifting from reactive maintenance to proactive preventive maintenance. By providing early warnings, this invention helps users take timely action, avoiding production interruptions and losses caused by sudden equipment failures, thereby improving industrial production efficiency.

[0056] 3. High computational efficiency, fully meeting the needs of industrial real-time monitoring: This invention combines the rapid classification capabilities of SVM (support vector machine) to quickly determine whether anomalies exist in real-time signal processing, greatly improving detection efficiency. By invoking LSTM for in-depth analysis only when abnormal trends are detected, this invention effectively reduces unnecessary computational burden, enabling the entire system to run efficiently on industrial sites and edge computing devices, meeting the needs of real-time monitoring.

[0057] 4. Broad Applicability and Strong Market Adaptability: This invention is widely applicable to industrial automation, electric vehicles, robotics, and other fields, and is suitable for monitoring the health of various types of motors, demonstrating strong market adaptability. By integrating into the Internet of Things (IoT) platform, this invention enables remote monitoring and intelligent maintenance, allowing users to monitor motor operating status anytime, anywhere and take timely action, significantly reducing operating costs for businesses.

[0058] In summary, this invention not only improves the accuracy and efficiency of motor anomaly detection but also enables fault trend prediction and preventive maintenance, promising broad market applications and significant economic benefits. This method, an efficient and intelligent motor health monitoring solution, is crucial for improving equipment safety and stability, reducing the risk of downtime, and increasing production efficiency.

[0059] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flow chart of an embodiment of a current fluctuation anomaly detection method based on machine learning LSTM and SVM of the present invention.

[0061] Figure 2 This is a schematic diagram of an LSTM training model in an embodiment of a current fluctuation anomaly detection method based on machine learning LSTM and SVM of the present invention. DETAILED DESCRIPTION

[0062] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the embodiments described are only some of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention.

[0063] References herein to "embodiments" mean that a particular feature, structure, or characteristic described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of this phrase in various places in the specification does not necessarily refer to the same embodiment, nor does it constitute an independent or alternative embodiment that is mutually exclusive of other embodiments. It is understood, both explicitly and implicitly, by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0064] See also Figure 1 This embodiment provides a method for detecting abnormal current fluctuations based on machine learning LSTM and SVM, which includes the following steps:

[0065] Step S1, data acquisition and preprocessing: collecting the motor's current signal and optional load signals, such as speed and torque, with a sampling frequency of 1kHz to 10kHz; preprocessing the collected signals;

[0066] Step S2, feature extraction: Calculate time domain features for determining load offset, detecting abnormal load, identifying abnormal load status, and detecting sudden faults; perform frequency domain feature extraction, obtain the signal's spectral features through fast Fourier transform, and use them to identify abnormal motor frequency and nonlinear distortion; calculate time series features, analyze the signal using short-time Fourier transform and wavelet transform, obtain the signal's frequency characteristics that change over time, and detect short-time impulse signals or sudden faults;

[0067] Step S3, machine learning model training: including SVM training and LSTM training, using the trained SVM model to judge the motor operating status in real time, and using the trained LSTM model to predict future current trends;

[0068] Step S4, abnormality detection and alarm:

[0069] SVM real-time classification: The real-time collected and pre-processed signals are input into the trained SVM model. If the SVM model determines that the motor operation status is abnormal, an alarm signal is output;

[0070] LSTM trend prediction: The calculated time series data is input into a trained LSTM model. If the LSTM model predicts abnormal current fluctuations in the future, an early warning signal is issued.

[0071] Step S5, remote monitoring: using the MQTT / HTTP protocol, the collected current signal, load signal, SVM classification results, and LSTM prediction results are uploaded to the cloud IoT platform, and remote data analysis is performed on the cloud IoT platform.

[0072] In the above step S1, the preprocessing includes:

[0073] The collected signals are subjected to noise reduction processing, using low-pass filtering to remove high-frequency noise and mean filtering to smooth the signals; the data are normalized to map the data to a specific range; and a sliding window method (e.g., 50 sampling points) is used to construct a time series data set;

[0074] in, I norm is the normalized value, I is the acquired current value, I min is the minimum current, I max The maximum current value is used to map the data to [0,1], which improves the model training results.

[0075] In the above step S2, the time domain features include:

[0076] Mean: Indicates the central trend of the signal and determines whether the load is offset. N is the number of sampling points, I i is the current value of the i-th sampling point.

[0077] Standard Deviation: Indicates the degree of signal fluctuation and detects abnormal loads.

[0078] Skewness: measures the symmetry of the signal distribution and identifies abnormal load conditions. If Skewness>0, it indicates that the distribution is right-skewed and the signal fluctuation may be biased towards the larger load. If Skewness<0, it indicates that the distribution is left-skewed and the signal fluctuation may be due to an asymmetric fault.

[0079] Kurtosis: measures whether the signal has prominent peaks and can be used to detect sudden faults. Kurtosis>3 indicates abnormal spikes, which may indicate a short-term overload or impact load. Kurtosis<3 indicates a relatively flat signal, which may indicate a stable load.

[0080] In the above step S2, the frequency spectrum characteristics of the signal are obtained by fast Fourier transform, including:

[0081] FFT transformation: extract the spectrum components of the current signal and identify abnormal motor frequencies. Among them F (k) is the amplitude of the kth frequency component, I(n) is the time domain signal, and N is the number of sampling points;

[0082] Harmonic analysis: Total harmonic distortion is used to measure the degree of nonlinear distortion of the current signal. F (1) is the fundamental wave, F (k) is the kth harmonic component.

[0083] In the above step S2, the signal is analyzed using short-time Fourier transform (STFT) and wavelet transform (WT), including:

[0084] Short-time Fourier transform STFT: analyzes the frequency characteristics of signals changing over time, Where ω(mn) is the window function, SIFT (n,k) is the k frequency components of the nth time segment;

[0085] Wavelet transform WT: detect short-time impulse signals or sudden faults, in is the mother wavelet function, a is the control scale (frequency), and b is the control translation (time).

[0086] In the above step S3, SVM training and LSTM training include:

[0087] SVM model training: Based on historical data, the motor operating status is labeled as normal or abnormal, such as normal (0) / abnormal (1). The labeled data is used to train the support vector machine SVM, and the Gaussian kernel function (RBF Kernel) is used for classification. K(x i ,x j )=exp(-γ||x i -x j || 2 ), and the SVM parameters (C, γ) are optimized through cross-validation method; after the training is completed, SVM is used to judge the motor operating status in real time.

[0088] LSTM model training: Construct time series data and use the sliding window method to construct the time series I(t+1)=f(I(t),I(t-1),…,I(tT)). Design the LSTM model structure based on the constructed time series features and use the mean square error (MSE) as the loss function to train the LSTM model:

[0089] Specifically, such as Figure 2 As shown, the LSTM model structure includes:

[0090] Input layer: receives T historical current data as input, where T is the preset number of historical data, and the historical current data is pre-processed motor current signal data;

[0091] Hidden layer: includes h layers of LSTM units. Each layer of LSTM units includes g neurons, where h is an integer greater than or equal to 1 and g is a positive integer. Each layer of LSTM units is connected in sequence to perform deep feature extraction and memory of the input historical current data.

[0092] Output layer: outputs the predicted current value at 1 to 5 time points in the future. The output layer calculates the predicted current value at different time points in the future through the fully connected layer based on the features extracted by the hidden layer.

[0093] In the above step S5, remote data analysis is performed on the cloud IoT platform, specifically including:

[0094] The uploaded data is stored and a historical database is established. Big data analysis algorithms are used to conduct in-depth mining of the stored data and analyze the correlation between current fluctuations and load changes. Visual reports are generated based on the analysis results to display trends and abnormalities in the motor's operating status.

[0095] In this embodiment, when uploading data to the cloud IoT platform, the transmitted data is encrypted using a hybrid encryption method:

[0096] First, the data is encrypted using the Advanced Encryption Standard AES algorithm;

[0097] The AES encryption key is then encrypted using the RSA algorithm;

[0098] Upload the encrypted data and the encrypted AES key to the cloud IoT platform;

[0099] On the cloud IoT platform, the RSA private key is first used to decrypt the AES encryption key, and then the decrypted AES encryption key is used to decrypt the data for remote data analysis.

[0100] In the above step S4, when using the SVM model, the calculated time domain features, frequency domain features and time series features are combined into an input feature vector, and the input feature vector is input into the trained support vector machine SVM model. The SVM model uses a Gaussian kernel function. The SVM model performs classification and judgment based on normal and abnormal labels pre-annotated based on historical data. If the output result is an abnormal category, the motor operation status is determined to be abnormal.

[0101] In the above step S4, a sliding window method is used to extract continuous time series segments from the preprocessed current signal data. Each segment contains T historical current data points as the input feature sequence of the LSTM model, where T is the length of the historical data set according to system requirements.

[0102] LSTM model input: The constructed time series data is input into a pre-trained LSTM model. The LSTM model has at least one hidden layer, each of which contains multiple LSTM units, which is used to learn the inherent laws and patterns of time series data.

[0103] Future current value prediction: The LSTM model predicts the current value at one or more future time points based on the input time series data to form a predicted current sequence.

[0104] Abnormal fluctuation detection: The predicted current sequence is compared with the preset current fluctuation threshold. If the predicted current value exceeds the threshold range, it is determined that the future current will experience abnormal fluctuations.

[0105] Warning signal is issued: When abnormal fluctuations in future current are detected, the system immediately issues a warning signal. The warning signal can be implemented through sound and light alarms, sending text messages or emails to designated user terminals, displaying warning information on the monitoring interface, etc., so that users can take timely measures to avoid potential faults.

[0106] In this embodiment, FFT and SVM can be used to trigger maintenance management to report bearing wear, time domain analysis and SVM can be used to immediately alarm for load mutations, wavelet transform and LSTM can be used to alarm for future trends, and LSTM can be used to alarm for overload operation 5 seconds in advance.

[0107] Bearing wear monitoring and alerting include using Fast Fourier Transform (FFT) to perform spectral analysis on current signals to extract specific frequency harmonic features associated with bearing wear. These extracted features are then fed into a trained support vector machine (SVM) model. If the model output indicates abnormal bearing wear, a maintenance alert is triggered, notifying personnel to inspect or replace the bearing.

[0108] Instant load change alarms: Time-domain analysis of the current signal calculates characteristics such as the degree of current change or rate of change. These characteristics are input into the SVM model. If the model determines that the load change is abnormal, an alarm is immediately issued, prompting the operator to pay attention and check the load condition.

[0109] Future Trend Warning: Wavelet transforms are used to analyze the current signal in the time and frequency domains, extracting signal details and trend characteristics. The wavelet transform results are combined with the predictions of the LSTM model. If abnormal fluctuations in the current signal are predicted in the future, an early warning signal is issued to support preventive maintenance decisions.

[0110] Overload early warning: An LSTM model performs deep learning on the time series data of current signals to predict current values ​​over a period of time. If the LSTM model predicts that the current value will exceed the preset overload threshold within the next 5 seconds, an overload alarm is issued 5 seconds in advance, allowing operators to take timely measures to prevent equipment damage.

[0111] It can be seen that through the above-mentioned diversified abnormality detection and processing strategies, the present invention can comprehensively and accurately monitor the operating status of the motor, and promptly issue alarms and maintenance notifications in different types of abnormal situations, thereby effectively ensuring the safe operation of the motor and reducing failure rates and maintenance costs.

[0112] Furthermore, the present invention employs a machine learning model for adaptive learning, automatically adjusting detection thresholds based on the motor's actual operating data under different operating conditions, effectively improving adaptability to complex operating conditions. By combining time-domain, frequency-domain, and time-frequency-domain feature analysis, the present invention can comprehensively and accurately detect various types of motor load anomalies, including but not limited to overload, bearing wear, and stalling, significantly reducing false positives and missed detections.

[0113] Furthermore, by leveraging the time series modeling capabilities of LSTM (Long Short-Term Memory) networks, this invention can deeply analyze historical current data and predict future trends, proactively identifying potential motor failures and shifting from reactive maintenance to proactive preventative maintenance. By providing early warnings, this invention helps users take timely action, avoiding production interruptions and losses caused by sudden equipment failures, thereby improving industrial production efficiency.

[0114] Furthermore, this invention combines the rapid classification capabilities of SVMs (support vector machines) to rapidly determine anomalies in real-time signal processing, significantly improving detection efficiency. By invoking LSTM for in-depth analysis only when anomalies are detected, this invention effectively reduces unnecessary computational burdens, enabling the entire system to operate efficiently in industrial sites and on edge computing devices, meeting the needs of real-time monitoring.

[0115] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0116] The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims

1. A current fluctuation anomaly detection method based on machine learning LSTM and SVM, characterized in that: The following steps are involved: Data acquisition and preprocessing: Collect the motor's current signal and load signal, and preprocess the collected signals; Feature extraction: Calculate time domain features to determine load offset, detect abnormal loads, identify abnormal load states, and detect sudden faults; perform frequency domain feature extraction and obtain the signal's spectral characteristics through fast Fourier transform to identify abnormal motor frequencies and nonlinear distortion levels; Calculate time series characteristics, analyze signals using short-time Fourier transform and wavelet transform, obtain the frequency characteristics of signals that change over time, and detect short-time impact signals or sudden faults; Machine learning model training: This includes support vector machine (SVM) training and long-term time-lapse (LSTM) training. The trained SVM model is used to determine the motor's operating status in real time, and the trained LSTM model is used to predict future current trends. Abnormal detection and alarm: SVM real-time classification: The real-time collected and pre-processed signals are input into the trained SVM model. If the SVM model determines that the motor operation status is abnormal, an alarm signal is output; LSTM trend prediction: The calculated time series data is input into a trained LSTM model. If the LSTM model predicts abnormal fluctuations in current in the future, an early warning signal is issued.

2. The method according to claim 1, characterized in that The pretreatment includes: The collected signals are subjected to noise reduction processing, low-pass filtering is used to remove high-frequency noise, and mean filtering is used to smooth the signals; the data are normalized to map the data to a specific range; and a sliding window method is used to construct a time series data set; in, I norm is the normalized value, I is the acquired current value, I min is the minimum current, I max is the maximum current value, so that the data is mapped to [0,1].

3. The method according to claim 1, characterized in that The time domain features include: Mean: Indicates the central trend of the signal and determines whether the load is offset. N is the number of sampling points, I i is the current value of the i-th sampling point; Standard Deviation: Indicates the degree of signal fluctuation and detects abnormal loads. Skewness: measures the symmetry of the signal distribution and identifies abnormal load conditions. If Skewness>0, it indicates that the distribution is right-skewed, and if Skewness<0, it indicates that the distribution is left-skewed; Kurtosis: measures whether the signal has prominent peaks and can be used to detect sudden faults. Kurtosis>3 indicates abnormal spikes, and Kurtosis<3 indicates a relatively flat signal.

4. The method according to claim 1, wherein The obtaining of the spectrum characteristics of the signal by fast Fourier transform includes: FFT transformation: extract the spectrum components of the current signal and identify abnormal motor frequencies. Among them F (k) is the amplitude of the kth frequency component, I(n) is the time domain signal, and N is the number of sampling points; Harmonic analysis: Total harmonic distortion is used to measure the degree of nonlinear distortion of the current signal. F (1) is the fundamental wave, F (k) is the kth harmonic component.

5. The method according to claim 1, wherein The signal is analyzed using short-time Fourier transform (STFT) and wavelet transform (WT), including: Short-time Fourier transform STFT: analyzes the frequency characteristics of signals changing over time, Where ω(mn) is the window function, SIFT (n,k) is the k frequency components of the nth time segment; Wavelet transform WT: detect short-time impulse signals or sudden faults, in is the mother wavelet function, a is the control scale, and b is the control translation.

6. The method according to claim 1, characterized in that The SVM training and LSTM training include: SVM model training: The motor operating status is labeled as normal or abnormal based on historical data. The labeled data is used to train the support vector machine (SVM). The Gaussian kernel function is used for classification, and the SVM parameters are optimized through cross-validation. LSTM model training: The LSTM model structure is designed based on the constructed time series features, and the mean square error (MSE) is used as the loss function to train the LSTM model.

7. The method according to claim 6, characterized in that The LSTM model structure includes: Input layer: receives T historical current data as input, where T is the preset number of historical data, and the historical current data is pre-processed motor current signal data; Hidden layer: includes h layers of LSTM units. Each layer of LSTM units includes g neurons, where h is an integer greater than or equal to 1 and g is a positive integer. Each layer of LSTM units is connected in sequence to perform deep feature extraction and memory of the input historical current data. Output layer: outputs the predicted current value at 1 to 5 time points in the future. The output layer calculates the predicted current value at different time points in the future through the fully connected layer based on the features extracted by the hidden layer.

8. The method according to any one of claims 1 to 7, characterized in that Also includes: Using the MQTT / HTTP protocol, the collected current signals, load signals, SVM classification results, and LSTM prediction results are uploaded to the cloud IoT platform for remote data analysis, including: The uploaded data is stored and a historical database is established. Big data analysis algorithms are used to conduct in-depth mining of the stored data and analyze the correlation between current fluctuations and load changes. Visual reports are generated based on the analysis results to display trends and abnormalities in the motor's operating status.

9. The method according to claim 8, characterized in that: When uploading data to the cloud IoT platform, the transmitted data is encrypted using a hybrid encryption method: First, the data is encrypted using the Advanced Encryption Standard AES algorithm; The AES encryption key is then encrypted using the RSA algorithm; Upload the encrypted data and the encrypted AES key to the cloud IoT platform; On the cloud IoT platform, the RSA private key is first used to decrypt the AES encryption key, and then the decrypted AES encryption key is used to decrypt the data for remote data analysis.

10. The method according to claim 6, wherein: When using the SVM model, the calculated time domain features, frequency domain features, and time series features are combined into an input feature vector, which is then input into the trained support vector machine (SVM) model. The SVM model uses a Gaussian kernel function and performs classification based on normal and abnormal labels pre-labeled based on historical data. If the output result is an abnormal category, the motor's operating status is determined to be abnormal.

Images