Neural network model-based abnormality diagnosis system and method for hydraulic movable arm of steel grabbing machine

Abstract

The invention discloses a steel grabbing machine hydraulic movable arm abnormity diagnosis system and method based on a neural network model, and particularly relates to the technical field of data processing. A high-sensitivity acoustic emission sensor and a multi-point temperature sensor are installed at the connecting position of a hydraulic system pipeline, acoustic signals and temperature data generated by hydraulic oil leakage and pipeline blockage are collected in real time, pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics are extracted, a data prediction model is constructed based on the extracted characteristics, and a data prediction result is obtained. The distinguishing accuracy of the neural network model on hydraulic oil leakage and pipeline blockage coupling faults is quantified, the condition that the model performance is insufficient is effectively recognized through accuracy division and abnormal degree analysis, the fault distinguishing capacity of the neural network is continuously optimized by updating a training set and retraining the model, and the fault distinguishing efficiency is improved. The precision and robustness of fault diagnosis of the hydraulic movable arm are remarkably improved, misjudgment and missed judgment are reduced, and the reliability of the system and the trust of operators are enhanced.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and in particular to a steel grabber hydraulic boom abnormality diagnosis system and method based on a neural network model. Background Art

[0002] Abnormal diagnosis of hydraulic boom of steel grabber based on neural network model refers to the use of neural network technology to monitor and analyze the operating status of hydraulic boom of steel grabber, aiming to identify and predict potential faults or abnormal conditions. The hydraulic boom of steel grabber is one of its core components, responsible for completing the process of grabbing, lifting and moving steel. Due to its long-term operation under high load and complex environment, it may suffer from wear, leakage or other abnormalities, affecting the operating efficiency and safety of the equipment.

[0003] The neural network model establishes nonlinear mapping relationships by learning the operating data of the hydraulic boom (such as pressure, flow, displacement, temperature, etc.), and can accurately detect abnormal behaviors and trends. This method has a high level of adaptability and intelligence, and can issue early warnings in the early stages of failures, thereby avoiding major equipment damage or production stagnation. Combined with big data analysis and real-time monitoring, the neural network-based diagnosis method improves equipment management efficiency and the level of intelligence in industrial production.

[0004] The prior art has the following deficiencies: Hydraulic boom anomalies may be caused by multiple faults occurring simultaneously, such as hydraulic oil leakage and pipeline blockage causing pressure anomalies. When dealing with complex multi-fault coupling phenomena, neural networks may not be able to accurately distinguish the characteristics of different faults, resulting in diagnostic failure or misjudgment. In addition, in the case of multi-fault coupling, if the diagnostic system repeatedly misjudges or misses the diagnosis, it may cause operators to gradually lose trust in the neural network diagnostic system. Summary of the invention

[0005] The purpose of the present invention is to provide a steel grabber hydraulic boom abnormality diagnosis system and method based on a neural network model to solve the shortcomings of the background technology.

[0006] In order to achieve the above object, the present invention provides the following technical solution: a method for diagnosing abnormality of a hydraulic boom of a steel grabber based on a neural network model, comprising the following steps: S1: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; S2: preprocessing the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracting the pulse density fluctuation characteristics in the preprocessed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; S3: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is constructed, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; S4: dividing the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performing corresponding processing; S5: For incomplete accuracy distinction, further analyze the abnormal degree of accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage. If the abnormal degree is high, update the training set of the neural network model and retrain it to improve the distinction accuracy.

[0007] Preferably, in S2, the pulse density fluctuation characteristics in the extracted acoustic wave signal are analyzed to generate a pulse density fluctuation index, and the pulse density fluctuation index is obtained by: The original sound wave signal collected in real time is segmented and windowed, and the expression is: ; x(t) is the original sound wave signal, is the short-time window of the window function shifted to time τ, is the windowed signal, which is used to intercept the signal data of the local period, perform short-time Fourier transform on the windowed signal, and calculate the expression: ; In the formula, is the complex-valued spectrum at time τ and frequency f, is the Fourier basis function, which is used to extract the signal components at frequency f and calculate the spectrum amplitude. It represents the intensity of the signal at time τ and frequency f, and the calculation expression is: ;in, yes The real part of yes The imaginary part of the frequency range to , the calculation expression of energy density is: ; Indicates that at time τ, to Energy density in the frequency range, Represents the energy spectrum at frequency f, calculates the pulse density fluctuation index, and the calculation expression is: ; In the formula, is the pulse density fluctuation index, M is the total number of windows, is the average density, indicating the overall energy level.

[0008] Preferably, in S2, the extracted temperature distribution uniformity deviation characteristics are analyzed to generate a temperature distribution uniformity deviation index, and the method for obtaining the temperature distribution uniformity deviation index is: Normalize the temperature data collected in real time from all monitoring points and calculate the expression: :In the formula, is the standardized temperature data, is the temperature data of the jth monitoring point at the i-th time point, is the mean temperature of the jth monitoring point, is the temperature standard deviation of the jth monitoring point; calculate the covariance matrix, the covariance matrix C represents the correlation of temperature changes between monitoring points, and the calculation expression is: ; In the formula, Z is the standardized temperature data matrix, the size is n×m, where n is the number of time points and m is the number of monitoring points. is the transposed matrix of Z, and the covariance matrix C is decomposed into features, and the expression is: ; is the eigenvalue of the kth principal component, indicating the variance contribution of the principal component, is the eigenvector corresponding to the kth eigenvalue, indicating the direction of the principal component, and all eigenvalues ​​are decomposed and the corresponding eigenvector , calculate the temperature distribution uniformity deviation index TDUEI, the expression is: ; In the formula, is the eigenvalue of the first principal component.

[0009] Preferably, in S3, a data prediction model is constructed based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model, specifically: The pulse density fluctuation index and the temperature distribution uniformity deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of all neural network models in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0010] Preferably, in S4, the accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage is divided into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, specifically: The obtained accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the first standard threshold and the second standard threshold respectively; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than the second standard threshold, it means that the neural network model has high accuracy in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage, and a high-accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is in the middle, and a medium accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is divided into incomplete accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is less than the first standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is low. At this time, a low accuracy distinction signal is generated, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as inaccurate distinction.

[0011] Preferably, in S5, for incomplete accuracy distinction, the abnormal degree of accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage is further analyzed. If the abnormal degree is high, the training set of the neural network model is updated and retrained, specifically: For incomplete accuracy distinction, that is, the accuracy value of the neural network model generated within a fixed time period to distinguish the coupling faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy abnormality degree is defined The calculation formula is: ; In the formula, is the degree of abnormal accuracy, indicating the prediction accuracy value Target accuracy value The deviation between is the accuracy value of the current prediction of the neural network model, is the expected target accuracy value under ideal working conditions preset based on historical data; The abnormality degree is divided into low abnormality degree: < ϵ1, no need for immediate retraining, medium abnormality: ϵ1≤ <ϵ2, check the data, high degree of anomaly: ≥ϵ2, the training set needs to be updated and the model needs to be retrained. ϵ1 and ϵ2 are the preset abnormality thresholds.

[0012] Preferably, the training set of the neural network model is updated, and the updating condition is: when When ≥ϵ2, the specific value is: The comprehensive feature vector of a single sample at the current moment And the corresponding true label Add to training set: ;in, is the output function of the model, The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage; To add new data, For the original data, new sample data is generated by perturbing the characteristics of the boundary conditions , the expression is: ; , is a small random perturbation, the new training set Defined as: ; For new sample data sets, the goal of retraining is to minimize the prediction error and optimize the accuracy value output of the neural network; Using the mean squared error loss function , the expression is: ; Where N is the number of samples in the updated training set, is the accuracy value of the model prediction, is the true accuracy value; the neural network weights are updated using gradient descent, and the expression is: ; In the formula, is the neural network weight of the sth iteration, η is the learning rate, is the gradient of the loss function to the weight, and the training is done until the loss function converges, that is: ; The preset loss convergence threshold uses accuracy to test the performance of the updated model.

[0013] The present invention also provides a steel grabber hydraulic boom abnormality diagnosis system based on a neural network model, comprising a data acquisition module, a feature extraction module, a prediction model construction module, an accuracy division module and an optimization module; Data acquisition module: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; Feature extraction module: pre-processes the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracts the pulse density fluctuation characteristics in the pre-processed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; Prediction model building module: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is built, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; Accuracy classification module: divides the accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performs corresponding processing; Optimization module: For incomplete accuracy distinction, further analysis is performed on the accuracy abnormality of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage. If the abnormality is high, the training set of the neural network model is updated and retrained to improve the distinction accuracy.

[0014] In the above technical solution, the technical effects and advantages provided by the present invention are: 1. The present invention introduces high-sensitivity acoustic emission sensors and multi-point temperature sensors, combines pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, accurately captures the complex coupling fault characteristics of hydraulic oil leakage and pipeline blockage in the hydraulic system, and constructs a data prediction model to quantitatively evaluate and grade the fault differentiation ability of the neural network model. For incomplete accuracy differentiation, through the analysis of the degree of abnormal accuracy, the dynamic update of the training set and the adaptive optimization of the model are realized, the fault diagnosis ability of the neural network under complex working conditions is improved, and the misjudgment and missed judgment are effectively reduced.

[0015] 2. The present invention can significantly improve the ability of the hydraulic boom abnormality diagnosis system to handle multiple fault coupling phenomena, achieve high accuracy and robustness of the diagnosis results, enhance operators' trust in the system, and ensure the safety and reliability of equipment operation. At the same time, by dynamically adjusting and optimizing the neural network model, the adaptability and intelligence level of the system in actual industrial scenarios are improved, providing an efficient, flexible and reliable solution for fault diagnosis of hydraulic systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0017] Figure 1 The figure is a flow chart of the method of the present invention.

[0018] Figure 2 It is a system module diagram of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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 creative work are within the scope of protection of the present invention.

[0020] Example 1, please refer to Figure 1 and Figure 2 As shown, the abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model described in this embodiment includes the following steps: S1: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; S2: preprocessing the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracting the pulse density fluctuation characteristics in the preprocessed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; S3: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is constructed, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; S4: dividing the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performing corresponding processing; S5: For incomplete accuracy distinction, further analyze the abnormal degree of accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage. If the abnormal degree is high, update the training set of the neural network model and retrain it to improve the distinction accuracy.

[0021] In S1, a high-sensitivity acoustic emission sensor is installed at the pipeline connection of the hydraulic system to collect real-time acoustic wave signals of different frequencies and intensities generated when hydraulic oil leaks and pipelines are blocked, and a multi-point temperature sensor is used to measure the temperature data of different monitoring points in the hydraulic system, specifically: The installation and configuration of the high-sensitivity acoustic emission sensor include: Frequency range: 50kHz to 500kHz, which can cover the acoustic signals generated by hydraulic oil leakage and pipeline blockage. Sensitivity: greater than 60dB, ensuring that weak signals can also be captured. Environmental adaptability: supports high temperature and high pressure environments (temperature range -20℃ to 120℃; pressure resistance ≥10MPa). Quantity and type: 3 to 5 sensors are selected according to the complexity of the pipeline and the working range of the hydraulic boom. Installation locations include: Pipeline connections: such as elbows and interfaces, where leakage signals are most likely to occur. High-pressure areas: such as pump outlets and valve inlets, where pipeline blockages may cause significant acoustic signals. Boom support points: near the force transmission part of the boom, used to capture acoustic signals transmitted by the structure. Installation methods include: Fixing method: Use a magnetic adsorption seat or high-temperature glue to fix the sensor to ensure stability and prevent vibration interference. Signal coupling: Apply a coupling agent (such as silicone or industrial grease) between the sensor and the pipeline surface to improve the efficiency of sound wave transmission. Use anti-interference shielded cables to transmit sensor signals to the data acquisition module. Equipped with a multi-channel high-frequency acquisition card (sampling rate ≥ 1MHz) to record sound wave signals in real time.

[0022] The installation and configuration of multi-point temperature sensors include: Parameter requirements: Measurement range: -40℃ to 150℃, covering the operating temperature range of the hydraulic system. Accuracy requirements: ±0.1℃, to ensure the accuracy of temperature difference analysis. Response time: less than 1 second, used to capture rapid temperature changes. Select high-pressure resistant thermocouples (such as K-type) or distributed fiber optic temperature sensors (DFOS) for layout. Installation locations include: Key monitoring points: Pump outlet: capture temperature changes in high-pressure areas. Valve connection: Valve failure may cause local temperature rise. Potential leakage areas: Hydraulic oil leakage is prone to occur at pipeline interfaces or boom ends. Cooling system inlet and outlet: monitor overall changes in temperature distribution. Installation methods include: Fixing method: Fix the thermocouple directly on the surface of the pipeline and protect it with metal straps and thermal insulation materials. Distributed fiber optic temperature sensors can be routed along the pipeline and fixed with high-temperature resistant glue. Thermal coupling treatment: Apply thermal conductive silicone grease between the sensor and the surface to improve temperature transfer efficiency. Optimize cable paths to avoid interference areas (such as high-voltage cables). Use protective sleeves to prevent mechanical damage. Equipped with a multi-channel temperature collector, it supports a resolution of at least 0.01°C and real-time monitoring.

[0023] Use a unified clock system to synchronize the acoustic emission signal and temperature data to ensure that the two data have the same time reference. Configure filters for the acoustic emission signal and temperature data respectively (such as high-pass filters to remove low-frequency noise). Automatically remove abnormal data points caused by equipment interference or sensor failure. Verify its sensitivity by artificially simulating leakage and blockage scenarios. Use a standard temperature source to calibrate the sensor measurement accuracy. After integrating the sensor, collector and data processing module, simulate the normal and fault conditions of the hydraulic system through pressure testing to verify the effectiveness of data acquisition and processing.

[0024] The collected data is transmitted to the central processing unit (CPU) using industrial bus (such as CAN bus or Modbus) or wireless transmission (such as LoRa or Zigbee). The real-time and reliability of the transmission process are ensured. The data analysis platform is used to display the real-time trend of acoustic emission signals and temperature gradients. Combined with the preset threshold range, the alarm mechanism of leakage or blockage is triggered.

[0025] S2: Preprocess the collected acoustic wave signals of different frequencies and intensities and temperature data, and extract the pulse density fluctuation characteristics in the preprocessed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data.

[0026] The preprocessing and feature extraction of acoustic wave signals include: using high-pass filters to remove low-frequency background noise, and separating different frequency components through wavelet transform to highlight the high-frequency characteristics of abnormal signals (such as continuous low-frequency signals of leakage and intermittent high-frequency signals of blockage). Extracting pulse density fluctuation characteristics, specifically including: Pulse counting: counting the number of pulses of acoustic wave signals per unit time, which is used to reflect the frequency of acoustic wave activities. Pulse density fluctuation amplitude: calculating the amplitude of the change in the number of pulses in different time windows, reflecting the dynamic characteristics of the signal. Frequency domain energy distribution: analyzing the energy distribution of the signal in a specific frequency band (such as 50-100kHz for leakage or 200-300kHz for blockage) to distinguish fault modes.

[0027] The preprocessing and feature extraction of temperature data include: noise filtering (such as Kalman filtering) of temperature data, eliminating outliers caused by sensor jitter or environmental fluctuations, and using the moving average method to smooth temperature changes. Extract the deviation characteristics of temperature distribution uniformity, including: Maximum temperature difference (ΔT_max): Calculate the temperature difference between the highest and lowest temperature monitoring points in the hydraulic system to reflect the uniformity of the overall temperature distribution. Temperature gradient (∇T): Calculate the local temperature change rate between different monitoring points to locate areas of abnormal temperature changes (such as cold leak points or hot blockage points).

[0028] The pulse density fluctuation characteristics in the extracted acoustic wave signal are analyzed to generate a pulse density fluctuation index. The method for obtaining the pulse density fluctuation index is as follows: The original sound wave signal collected in real time is segmented and windowed, and the expression is: ; x(t) is the original sound wave signal, It is a short-time window whose window function is shifted to time τ. Hamming window or rectangular window is commonly used. is the windowed signal, which is used to intercept the signal data of the local period, perform short-time Fourier transform on the windowed signal, and calculate the expression: ; In the formula, is the complex-valued spectrum at time τ and frequency f, is the Fourier basis function, which is used to extract the signal components at frequency f and calculate the spectrum amplitude. It represents the intensity of the signal at time τ and frequency f, and the calculation expression is: ;in, yes The real part of yes The imaginary part of the frequency range to , such as the low frequency band (50–100kHz) of hydraulic oil leakage signals or the high frequency band (200–300kHz) of pipeline blockage. The calculation expression of energy density is: ; Indicates that at time τ, to Energy density in the frequency range, Represents the energy spectrum at frequency f, calculates the pulse density fluctuation index, and the calculation expression is: ; In the formula, is the pulse density fluctuation index, M is the total number of windows, is the average density, indicating the overall energy level.

[0029] The larger the pulse density fluctuation index (PDFI), the stronger the volatility of the acoustic signal and the more dramatic the change in energy density over time. This situation is often related to the characteristics of pipeline blockage, because the acoustic signal caused by blockage is usually a high-frequency, intermittent pulse signal, which leads to a significant increase in the fluctuation amplitude of the pulse density. When the PDFI is large, the neural network model can more easily capture the characteristics of pipeline blockage and distinguish coupling faults with higher accuracy. However, if the PDFI exceeds the normal range too much, it may also mean that there is noise interference or false detection in the data, affecting the reliability of the model judgment.

[0030] On the contrary, the smaller the PDFI, the weaker the volatility of the acoustic signal, and the more stable the change of energy density over time. This situation is often related to the characteristics of hydraulic oil leakage, because the acoustic signal caused by the leakage is usually low-frequency and continuous, resulting in a low amplitude of pulse density fluctuation. When the PDFI is small, the neural network model can effectively identify the leakage characteristics, thereby improving the accuracy of distinguishing the leakage part in the coupling fault. However, if the PDFI is too small, it may mean that the signal energy is insufficient or the key features are not captured, which may reduce the overall diagnostic ability of the model for complex coupling faults.

[0031] After analyzing the extracted temperature distribution uniformity deviation characteristics, a temperature distribution uniformity deviation index is generated. The method for obtaining the temperature distribution uniformity deviation index is as follows: Normalize the temperature data collected in real time from all monitoring points to make its mean 0 and standard deviation 1. Calculate the expression: :In the formula, is the standardized temperature data, is the temperature data of the jth monitoring point at the i-th time point, is the mean temperature of the jth monitoring point, is the temperature standard deviation of the jth monitoring point; calculate the covariance matrix, the covariance matrix C represents the correlation of temperature changes between monitoring points, and is used to extract the main change mode. The calculation expression is: ; In the formula, Z is the standardized temperature data matrix, the size is n×m, where n is the number of time points and m is the number of monitoring points. is the transposed matrix of Z, and the covariance matrix C is decomposed into features, and the expression is: ; is the eigenvalue of the kth principal component, indicating the variance contribution of the principal component, is the eigenvector corresponding to the kth eigenvalue, indicating the direction of the principal component, and all eigenvalues ​​are decomposed (in descending order) and the corresponding eigenvectors , calculate the temperature distribution uniformity deviation index TDUEI, the expression is: ; In the formula, is the eigenvalue of the first principal component, indicating the variance contribution of the main variation mode.

[0032] The larger the temperature distribution uniformity deviation index (TDUEI), the more uneven the temperature distribution of the hydraulic system. This usually means that there is a significant local temperature anomaly, such as high temperature accumulation caused by pipe blockage. In this case, the neural network model has a more obvious basis for distinguishing the characteristics of pipe blockage, thereby improving the diagnostic accuracy of the blocked part of the coupling fault. However, if the TDUEI is too large, it may also indicate that there is abnormal noise or other complex interference factors in the data, which may cause the neural network to misjudge the overall diagnosis of the coupling fault.

[0033] On the contrary, the smaller the TDUEI, the more uniform the temperature distribution of the hydraulic system and the less obvious local temperature anomalies. This may be related to hydraulic oil leakage, because leakage usually causes a gentler temperature gradient change. At this time, the neural network model is more likely to capture the leakage characteristics, thereby improving the diagnostic accuracy of the leakage part of the coupling fault. However, if the TDUEI is too small, it may reflect that the change amplitude of the temperature data is insufficient, resulting in a decrease in the neural network's ability to distinguish complex coupling faults, especially when it is difficult to detect obvious blockage features.

[0034] S3: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is constructed, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model.

[0035] The pulse density fluctuation index and the temperature distribution uniformity deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of all neural network models in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

[0036] The method for obtaining the accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is as follows: from the comprehensive feature vector training data of the trained machine learning model, the corresponding function expression is obtained: ; In the formula, is the output function of the model, PDFI is the pulse density fluctuation index, TDUEI is the temperature distribution uniformity deviation index, The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage.

[0037] S4: The accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is divided into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and corresponding processing is performed.

[0038] The obtained accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the first standard threshold and the second standard threshold respectively; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than the second standard threshold, it means that the neural network model has high accuracy in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage, and a high-accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is in the middle, and a medium accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is divided into incomplete accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is less than the first standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is low. At this time, a low accuracy distinction signal is generated, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as inaccurate distinction.

[0039] When the neural network model's ability to distinguish reaches a high accuracy, it means that the model can effectively distinguish the coupled faults of hydraulic oil leakage and pipeline blockage under the current working conditions. At this time, the main task is to maintain the model performance and accumulate data for subsequent optimization and evaluation.

[0040] Record the prediction results of the current model and the corresponding comprehensive feature vector (i.e., PDFI and TDUEI values). Record the operating parameters and environmental characteristics of the system (such as temperature, pressure, and noise level) to provide a basis for subsequent analysis. Save the current model parameters as a benchmark and establish a version control mechanism for subsequent comparison and optimization. Continuously monitor new data input to verify the stability of model performance. If high accuracy is maintained for a long time, the frequency of model updates can be reduced, reducing computing and maintenance costs. Add historical data under high accuracy to the model evaluation library as a control sample under other working conditions.

[0041] Medium accuracy indicates that the neural network model can distinguish between hydraulic oil leakage and pipeline blockage to a certain extent, but the distinction effect is insufficient. This may be due to the increased complexity of the specific working condition or the model's insufficient adaptability to certain features. At this time, it is necessary to deeply analyze the cause of the problem and optimize the model. Analyze the comprehensive feature vector to identify specific value ranges in PDFI and TDUEI (such as areas close to the threshold). Check whether the input data is abnormal or insufficient (such as missing samples for specific working conditions). Increase sample data under complex working conditions (such as feature data when leakage and blockage occur at the same time). Use data enhancement methods to generate simulated data, such as noise injection or boundary value interpolation. Add feature engineering to mine other implicit features that may affect accuracy (such as flow fluctuations or pressure change rate). Adjust the hyperparameters of the neural network model (such as learning rate, number of layers) or introduce an attention mechanism to strengthen the model's attention to key features. After the model is optimized, test the new working condition data to evaluate whether the accuracy is improved. If the high accuracy state is still not achieved after optimization, consider alternative algorithms (such as hybrid models). In the medium accuracy state, generate a prompt signal to alert the operator, but do not trigger emergency intervention. Collect more operating data to supplement the model training set.

[0042] Low accuracy indicates that the neural network model cannot effectively distinguish the coupled faults of hydraulic oil leakage and pipeline blockage under the current working conditions. This may be due to unclear feature distribution, insufficient data, or overfitting / underfitting of the model. At this time, emergency treatment measures need to be taken to restore the diagnostic ability of the model. Generate a low-accuracy differentiation signal to indicate potential system failure and notify the operator to intervene. Enable alternative fault diagnosis methods (such as rule-based traditional diagnosis systems) for temporary diagnosis. Check the abnormality of the comprehensive feature vector: whether there are cases where the feature value exceeds the reasonable range. Check the collection integrity and quality of sensor data to eliminate sensor failure or environmental interference. Analyze the difference between training data and actual data to identify data distribution deviation. Re-collect or generate targeted data: increase abnormal working condition data, especially boundary conditions (such as extreme leakage or blockage). Obtain more complex coupled fault samples through simulation experiments. Fully retrain the neural network model to ensure that the model can adapt to complex scenarios. Increase model complexity (such as deeper network layers or more feature fusion). Introduce ensemble learning methods (such as random forest or XGBoost) in combination with neural networks to enhance diagnostic capabilities. Use transfer learning to quickly improve accuracy with pre-trained models from other similar operating conditions. After completing model optimization, perform a comprehensive system calibration to ensure that diagnostic accuracy is restored. Test the model again under different operating conditions and adjust the standard threshold to better suit the actual scenario.

[0043] S5: For incomplete accuracy distinction, further analyze the abnormal degree of accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage. If the abnormal degree is high, update the training set of the neural network model and retrain it to improve the distinction accuracy.

[0044] For incomplete accuracy distinction, that is, the accuracy value of the neural network model generated within a fixed time period to distinguish the coupling faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy abnormality degree is defined The calculation formula is: ; In the formula, is the degree of abnormal accuracy, indicating the prediction accuracy value Target accuracy value The relative deviation between is the accuracy value of the current prediction of the neural network model, It is the expected target accuracy value under ideal working conditions preset based on historical data.

[0045] The abnormality degree is divided into low abnormality degree: < ϵ1, no need to retrain immediately. Moderate abnormality: ϵ1≤ <ϵ2, it is recommended to check the data and observe the performance trend, high abnormality level: ≥ϵ2, the training set needs to be updated and the model needs to be retrained. ϵ1 and ϵ2 are preset abnormality thresholds, which are usually adjusted by experimental data.

[0046] Update the training set of the neural network model. The update condition is: When ≥ϵ2, it is considered that the model’s current ability to distinguish coupling faults is insufficient and needs to be improved by updating the training set.

[0047] The comprehensive feature vector of a single sample at the current moment And the corresponding true label (Human judgment or backup system output) Add to training set: ; To add new data, The original data is generated by perturbing the features under boundary conditions or noise environment to generate new sample data. , the expression is: ; , is a small random perturbation used to simulate diverse scenarios. Updated training set: New training set Defined as: ; For the new sample data set, the goal of retraining is to minimize the prediction error and optimize the accuracy value output of the neural network. The mean square error (MSE) loss function is used. , the expression is: ; Where N is the number of samples in the updated training set, is the accuracy value of the model prediction, is the actual accuracy value.

[0048] Use the gradient descent method to update the neural network weights, the expression is: ; In the formula, is the neural network weight of the sth iteration, η is the learning rate, is the gradient of the loss function to the weight, and the training is done until the loss function converges, that is: ; The preset loss convergence threshold. Use accuracy to test the performance of the updated model. The expression is: Accuracy = number of correctly predicted samples / total number of test samples.

[0049] In this embodiment, by installing high-sensitivity acoustic emission sensors and multi-point temperature sensors, the acoustic wave signals and temperature data generated by hydraulic oil leakage and pipeline blockage in the hydraulic system are collected in real time, and the collected data is preprocessed to extract pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics. A data prediction model is constructed based on the extracted features to evaluate the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage, and it is divided into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and corresponding processing is performed respectively. For incomplete accuracy distinction, the degree of accuracy abnormality is further analyzed. If the degree of abnormality is high, the training set is updated and the neural network model is retrained to optimize its fault distinction ability and ensure the accuracy and reliability of hydraulic system fault diagnosis.

[0050] Embodiment 2, the abnormal diagnosis system of the hydraulic boom of a steel grabber based on a neural network model described in this embodiment includes a data acquisition module, a feature extraction module, a prediction model construction module, an accuracy division module and an optimization module; Data acquisition module: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; Feature extraction module: pre-processes the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracts the pulse density fluctuation characteristics in the pre-processed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; Prediction model building module: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is built, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; Accuracy classification module: divides the accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performs corresponding processing; Optimization module: For incomplete accuracy distinction, further analysis is performed on the accuracy abnormality of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage. If the abnormality is high, the training set of the neural network model is updated and retrained to improve the distinction accuracy.

[0051] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be covered by the protection scope of the present application.

Claims

1. A method for diagnosing abnormality of hydraulic boom of steel grabber based on neural network model, characterized in that: The following steps are involved: S1: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; S2: preprocessing the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracting the pulse density fluctuation characteristics in the preprocessed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; S3: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is constructed, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; S4: dividing the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performing corresponding processing; S5: For incomplete accuracy distinction, further analyze the abnormal degree of accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage. If the abnormal degree is high, update the training set of the neural network model and retrain it to improve the distinction accuracy.

2. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 1 is characterized in that: In S2, the pulse density fluctuation characteristics in the extracted acoustic wave signal are analyzed to generate a pulse density fluctuation index. The pulse density fluctuation index is obtained as follows: The original sound wave signal collected in real time is segmented and windowed, and the expression is: ; x(t) is the original sound wave signal, is the short-time window of the window function shifted to time τ, is the windowed signal, which is used to intercept the signal data of the local period, perform short-time Fourier transform on the windowed signal, and calculate the expression: ; In the formula, is the complex-valued spectrum at time τ and frequency f, is the Fourier basis function, which is used to extract the signal components at frequency f and calculate the spectrum amplitude. It represents the intensity of the signal at time τ and frequency f, and the calculation expression is: ;in, yes The real part of yes The imaginary part of the frequency range to , the calculation expression of energy density is: ; Indicates that at time τ, to Energy density in the frequency range, Represents the energy spectrum at frequency f, calculates the pulse density fluctuation index, and the calculation expression is: ; In the formula, is the pulse density fluctuation index, M is the total number of windows, is the average density, indicating the overall energy level.

3. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 2 is characterized in that: In S2, the extracted temperature distribution uniformity deviation characteristics are analyzed to generate a temperature distribution uniformity deviation index. The method for obtaining the temperature distribution uniformity deviation index is: Normalize the temperature data collected in real time from all monitoring points and calculate the expression: :In the formula, is the standardized temperature data, is the temperature data of the jth monitoring point at the i-th time point, is the mean temperature of the jth monitoring point, is the temperature standard deviation of the jth monitoring point; calculate the covariance matrix, the covariance matrix C represents the correlation of temperature changes between monitoring points, and the calculation expression is: ; In the formula, Z is the standardized temperature data matrix, the size is n×m, where n is the number of time points and m is the number of monitoring points. is the transposed matrix of Z, and the covariance matrix C is decomposed into features, and the expression is: ; is the eigenvalue of the kth principal component, indicating the variance contribution of the principal component, is the eigenvector corresponding to the kth eigenvalue, indicating the direction of the principal component, and all eigenvalues ​​are decomposed and the corresponding eigenvector , calculate the temperature distribution uniformity deviation index TDUEI, the expression is: ; In the formula, is the eigenvalue of the first principal component.

4. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 3 is characterized in that: In S3, a data prediction model is constructed based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics. The accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model. Specifically: The pulse density fluctuation index and the temperature distribution uniformity deviation index are converted into comprehensive feature vectors, and the comprehensive feature vectors are used as the input of the machine learning model. The machine learning model uses each group of comprehensive feature vectors to predict the accuracy value label of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the prediction target, and takes minimizing the sum of the prediction errors of the accuracy value labels of all neural network models in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage as the training target. The machine learning model is trained until the sum of the prediction errors converges and the model training is stopped. The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined according to the model output results, wherein the machine learning model is a polynomial regression model.

5. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 4 is characterized in that: In S4, the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is divided into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, specifically: The obtained accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the gradient standard threshold, the gradient standard threshold includes a first standard threshold and a second standard threshold, and the first standard threshold is less than the second standard threshold, and the accuracy value of the neural network model for distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is compared with the first standard threshold and the second standard threshold respectively; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than the second standard threshold, it means that the neural network model has high accuracy in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage, and a high-accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is in the middle, and a medium accuracy distinction signal is generated at this time, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is divided into incomplete accuracy distinction; If the accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is less than the first standard threshold, it means that the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is low. At this time, a low accuracy distinction signal is generated, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is classified as inaccurate distinction.

6. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 5 is characterized in that: In S5, for the incomplete accuracy distinction, the abnormal degree of accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage is further analyzed. If the abnormal degree is high, the training set of the neural network model is updated and retrained, specifically: For incomplete accuracy distinction, that is, the accuracy value of the neural network model generated within a fixed time period to distinguish the coupling faults of hydraulic oil leakage and pipeline blockage is greater than or equal to the first standard threshold and less than or equal to the second standard threshold, the accuracy abnormality degree is defined The calculation formula is: ; In the formula, is the degree of abnormal accuracy, indicating the prediction accuracy value Target accuracy value The deviation between is the accuracy value of the current prediction of the neural network model, is the expected target accuracy value under ideal working conditions preset based on historical data; The abnormality degree is divided into low abnormality degree: < ϵ1, no need to retrain immediately, medium abnormality: ϵ1≤ <ϵ2, check the data, high degree of anomaly: ≥ϵ2, the training set needs to be updated and the model needs to be retrained. ϵ1 and ϵ2 are the preset abnormality thresholds.

7. The abnormality diagnosis method of the hydraulic boom of a steel grabber based on a neural network model according to claim 6 is characterized by: Update the training set of the neural network model. The update condition is: When ≥ϵ2, the specific value is: The comprehensive feature vector of a single sample at the current moment And the corresponding true label Add to training set: ;in, is the output function of the model, The accuracy value of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage; To add new data, For the original data, new sample data is generated by perturbing the characteristics of the boundary conditions , the expression is: ; , is a small random perturbation, the new training set Defined as: ; For new sample data sets, the goal of retraining is to minimize the prediction error and optimize the accuracy value output of the neural network; Using the mean squared error loss function , the expression is: ; Where N is the number of samples in the updated training set, is the accuracy value of the model prediction, is the true accuracy value; the neural network weights are updated using gradient descent, and the expression is: ; In the formula, is the neural network weight of the sth iteration, η is the learning rate, is the gradient of the loss function to the weight, and the training is done until the loss function converges, that is: ; The preset loss convergence threshold uses accuracy to test the performance of the updated model.

8. A steel grabber hydraulic boom abnormality diagnosis system based on a neural network model, used to implement a steel grabber hydraulic boom abnormality diagnosis method based on a neural network model as claimed in any one of claims 1 to 7, characterized in that: It includes data collection module, feature extraction module, prediction model building module, accuracy division module and optimization module; Data acquisition module: Install high-sensitivity acoustic emission sensors at the pipe connections of the hydraulic system to collect real-time acoustic signals of different frequencies and intensities generated when hydraulic oil leaks and pipes are blocked, and use multi-point temperature sensors to measure temperature data at different monitoring points in the hydraulic system; Feature extraction module: pre-processes the collected acoustic wave signals of different frequencies and intensities and temperature data, and extracts the pulse density fluctuation characteristics in the pre-processed acoustic wave signals and the temperature distribution uniformity deviation characteristics in the temperature data; Prediction model building module: Based on the extracted pulse density fluctuation characteristics and temperature distribution uniformity deviation characteristics, a data prediction model is built, and the accuracy of the neural network model in distinguishing the coupled faults of hydraulic oil leakage and pipeline blockage is determined based on the output results of the prediction model; Accuracy classification module: divides the accuracy of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage into accuracy distinction, incomplete accuracy distinction and inaccuracy distinction, and performs corresponding processing; Optimization module: For incomplete accuracy distinction, further analysis is performed on the accuracy abnormality of the neural network model in distinguishing the coupling faults of hydraulic oil leakage and pipeline blockage. If the abnormality is high, the training set of the neural network model is updated and retrained to improve the distinction accuracy.

Images