Hydraulic gear pump assembly detection system

Abstract

The invention discloses a hydraulic gear pump assembly detection system, and relates to the technical field of product quality inspection. The hydraulic gear pump assembly detection system comprises a prediction model building module, an actual transmission comparison module and a multi-working-condition expansion analysis module. Through early fusion of multi-modal data and self-adaptive feature extraction, an accurate prediction model is constructed, potential anomalies can be effectively identified in advance, a prediction data set or an anomaly report is output in advance, and the detection efficiency is improved; the gear pump is driven to run at a low speed, predicted and actual gap data are further compared, deviation is calculated, an abnormal angle is marked, and an error source is analyzed, so that problem positioning is more accurate; and finally, considering the influence of speed and temperature, constructing a digital twin model to simulate different working conditions, positioning an error source in combination with multi-dimensional data, and outputting optimization suggestions, thereby improving the assembly efficiency. And dynamic detection is carried out and multi-parameter analysis is combined, so that anomaly analysis is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of product quality inspection, and specifically to a hydraulic gear pump assembly detection system. Background Art

[0002] Publication No. CN118735139B discloses a hydraulic gear pump assembly detection system, which relates to the field of hydraulic gear pump assembly detection. The hydraulic gear pump assembly detection system includes: a data acquisition subsystem for acquiring component angle data, component distance data, and component surface data of the assembled hydraulic gear pump; a data analysis subsystem for analyzing the component angle data, component distance data, and component surface data of the assembled hydraulic gear pump respectively to obtain angle detection indexes, linear dimension detection indexes, and surface detection indexes of the assembled hydraulic gear pump. By integrating a variety of detection modules and technical means, the invention can comprehensively evaluate the assembly quality of the hydraulic gear pump, not limited to a single quality index, but covering multiple aspects such as angle, dimension, and surface, thereby significantly improving the accuracy and systematicness of detection and ensuring that each assembly link meets high-quality standards.

[0003] However, as shown by this technology, the current hydraulic gear pumps are generally detected for the size data of each part directly in a static state after assembly to judge part errors, assembly errors, etc. However, in a static state, it is difficult to comprehensively detect the error data at each operating angle, even at different speeds and temperatures with a single angle, which is related to whether it can operate normally during actual operation. Although the operating state of the finished product will be sampled and inspected later, it cannot be targeted at each device, resulting in missed inspection problems. Moreover, finding the cause after a problem occurs greatly reduces the efficiency. If comprehensive detection can be carried out at the assembly stage, the efficiency and quality will be effectively improved. Therefore, the existing technology still has deficiencies in the quality inspection of hydraulic gear pumps after assembly. Summary of the Invention

[0004] In view of the deficiencies of the prior art, the present invention provides a hydraulic gear pump assembly detection system, which solves the problems mentioned above.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A hydraulic gear pump assembly detection system includes: Prediction model establishment module: Collect multi-modal data of images, vibrations, and pressures and perform preprocessing; adopt an early fusion strategy to fuse images, vibration signals, and pressure data at the feature level; adaptively extract gear tooth profile and clearance width features, and introduce an attention mechanism to automatically focus on the area most sensitive to clearance changes; construct a mathematical model to predict clearance changes, and perform model training and optimization; calculate the change rate of the predicted clearance, identify potential anomalies, and output a predicted clearance data set or an anomaly report; Actual transmission comparison module: Drive the gear pump to run at low speed, synchronously record images, timestamps, and angle information, and use 3D reconstruction technology to construct a 3D model of the gear pump; compare the predicted data with the clearance data in the 3D model; calculate the mean and standard deviation of the gear clearance, determine the fluctuation value, and identify periodic characteristics; compare the actual clearance data with the predicted value, calculate the deviation, mark abnormal angles, and analyze the error sources. Multi-condition expansion analysis module: Collect data at different speeds, analyze the influence of speed on clearance fluctuation and jitter frequency; monitor the working temperature of the gear pump, analyze the influence of thermal expansion on the clearance, and correct the parameters of the prediction model; construct a digital twin model of the gear pump to simulate the operating state under different conditions; introduce a reinforcement learning algorithm to optimize the assembly process parameters; combine multi-dimensional data to locate the main error sources and output assembly optimization suggestions.

[0006] Preferably, the content of multi-modal data collection and fusion performed by the prediction model establishment module includes: Data collection: Use a high-resolution industrial camera to collect images of the assembled gear pump, covering the full angle of 0° to 360°, and define the angle step B, and collect a total of images; Integrate vibration sensors and pressure sensors to synchronously collect the vibration signal V(t) and pressure data P(t) during the operation of the gear pump. Data preprocessing: Perform preprocessing steps of denoising, grayscale conversion, and edge enhancement on the images to extract gear contour and clearance features; Perform filtering and normalization on the vibration signal and pressure data to remove noise and outliers; Multi-modal fusion: Adopt an early fusion strategy, use a multi-modal fusion algorithm to fuse the preprocessed image data I, vibration signal V, and pressure data P at the feature level. The fused feature vector F is expressed as: F = Fusion(I, V, P) = W I ·f I (I) + W V ·f V (V) + W P ·f P (P); Among them, f I , f V , f P are the feature extraction functions of the image, vibration signal, and pressure data respectively, and W I , W V , W P are weight coefficients.

[0007] Preferably, the content of the adaptive feature extraction performed by the prediction model establishment module includes: Feature extraction: Extract key parameters of gear tooth profile and clearance width based on deep learning or traditional CV algorithms; Introduce an attention mechanism to automatically focus on the area in the image that is most sensitive to clearance changes. The attention weight A is calculated as: A = Softmax(W a ·h); where h is the feature vector, and W a is the weight matrix of the attention mechanism. The Softmax function is used to calculate the attention weight A to ensure that the sum of all weights is 1, thereby representing the relative probability of different regions' sensitivity to clearance changes; Adaptive adjustment: Automatically adjust the parameters and strategies of feature extraction according to the different working states and image characteristics of the gear pump. Use the adaptive learning rate adjustment algorithm, and the learning rate η is dynamically adjusted according to the gradient magnitude: ; where η0 is the initial learning rate, β1 and β2 are momentum parameters, m t , v t are the first-order and second-order momenta respectively, is a small constant to prevent division by zero.

[0008] Preferably, the content of the prediction model establishment module for constructing a mathematical model and identifying potential anomalies includes: Model selection: Construct a dynamic transmission prediction model. The input is the angle sequence θ = [θ1, θ2,..., θ N and the fused multi-modal feature F, and the output is the predicted clearance value corresponding to the angle ; Model training: Use historical data to train the prediction model and optimize the model parameters. The loss function L can be defined as the mean square error: ; where g i is the actual clearance value; Rate of change calculation: Calculate the rate of change r i of the predicted clearance with respect to the angle: ; Set the rate of change mutation threshold T r . If the rate of change mutation ∣r i ∣ > T r , it is determined that there is a potential anomaly; Anomaly prediction: If there is no anomaly, output the predicted clearance data set; if there is an anomaly, mark the abnormal angle and output an anomaly report.

[0009] Preferably, the content of the actual transmission comparison module for collecting images and constructing a three-dimensional model at low speed includes: Drive and collection: Drive the gear pump to run at low speed, set the number of frames to be collected, and synchronously record the time stamp t i and the angle information θ i ; Image grouping: Group by angle and extract the image set I of the same angle θ ; Three-dimensional reconstruction: Use three-dimensional reconstruction technology to reconstruct the collected two-dimensional images into a three-dimensional model M of the gear pump.

[0010] Preferably, the content of the actual transmission comparison module for comparative analysis based on the three-dimensional model and actual data includes: Comparison model analysis: Compare the clearance data g in the three-dimensional model M and the predicted data , calculate the deviation e i : ; Through the visualization tool, visually display the assembly error and clearance distribution; Clearance fluctuation calculation: For the image set I of the same angle θ , calculate the mean value μ of the gear clearance g and the standard deviation σ g : ; ; Combine the mean value μ g , the standard deviation σ g , engineering experience and design requirements to determine the fluctuation value; Periodic feature recognition: Through Fourier transform or wavelet analysis, identify the periodic feature f of the clearance fluctuation p , and correlate with the gear jitter situation; Deviation calculation for each angle: Compare the actual clearance data g with the predicted value angle by angle, and calculate the deviation e i ; Abnormal marking and analysis: Set the deviation threshold. If the deviation exceeds the threshold, mark it as an abnormal angle, and analyze the error source in combination with the jitter characteristics.

[0011] Preferably, the analysis of the speed sensitivity of the multi-condition expansion analysis module includes: Data collection: Repeat the collection and analysis process under low-speed transmission, and continue to collect data at medium speed and high speed; Influence analysis: Analyze the influence of the speed v on the standard deviation σ of the clearance fluctuation g , the jitter frequency fd Establish a speed-error correlation model based on the influence of σ g = a·v + b; f d = c·v d ; where a, b, c, and d are model parameters.

[0012] Preferably, the analysis of temperature sensitivity by the multi-condition expansion analysis module includes: Temperature monitoring: Integrate a temperature sensor to monitor the working temperature T of the gear pump; Data comparison: Repeat image acquisition and comparison at different temperatures to analyze the influence of thermal expansion on the clearance g; The coefficient of thermal expansion α can be expressed as: ; where Δg is the change in clearance, ΔT is the change in temperature, and g0 is the initial clearance; Model correction: According to the temperature influence, correct the prediction model parameters to improve the accuracy of the model at different temperatures.

[0013] Preferably, the content of constructing a digital twin model by the multi-condition expansion analysis module includes: Model construction: Construct a digital twin model M' of the gear pump and map the actually collected data to the digital twin model in real time; Simulation and verification: Predict the performance and reliability by simulating the operating state of the gear pump under different conditions; And compare and verify with the actual detection results to optimize the assembly process.

[0014] Preferably, the content of assembly optimization based on reinforcement learning by the multi-condition expansion analysis module includes: Agent definition: Introduce a reinforcement learning algorithm, take the assembly process parameters p as the actions of the agent, and take the detection result r as the reward signal; Optimization process: Continuously try different combinations of assembly parameters to let the agent learn the assembly process; Set the reward function R. The detection result r includes the deviation and jitter frequency of the clearance, then the calculation formula of R is: ; where λ is the weight coefficient to balance the influence of clearance deviation and jitter frequency; Gradually optimize the assembly parameters: Use the policy gradient algorithm to update the assembly parameters to improve the quality and reliability of the gear pump; Data analysis: Combine three-dimensional data of angle, speed, and temperature, and locate the main error sources through PCA or decision tree algorithms; Optimization suggestions: Output assembly optimization suggestions in combination with historical experience.

[0015] The present invention provides a hydraulic gear pump assembly detection system. Compared with the prior art, it has the following beneficial effects: 1. For this hydraulic gear pump assembly detection system, through early fusion of multi-modal data and adaptive feature extraction, combined with the attention mechanism to focus on sensitive areas, a precise prediction model is constructed, which can effectively identify potential anomalies in advance, output the prediction data set or anomaly report in advance, and improve the detection efficiency; the actual transmission comparison module drives the gear pump to run at low speed, constructs a model using 3D reconstruction technology, further compares the predicted and actual clearance data, calculates the deviation, marks the abnormal angle and analyzes the error source, making the problem positioning more accurate; finally, the multi-condition extended analysis module considers the influence of speed and temperature, corrects the model parameters, constructs a digital twin model to simulate different working conditions, introduces a reinforcement learning algorithm to optimize the assembly process parameters, combines multi-dimensional data to locate the error source and outputs optimization suggestions, improving the assembly efficiency; and dynamic detection is carried out and combined with multi-parameter analysis, and the anomaly analysis is more accurate.

[0016] 2. For this hydraulic gear pump assembly detection system, a high-resolution industrial camera collects images at all angles, combined with vibration and pressure sensors to synchronously collect data, providing a rich and accurate information source for subsequent analysis; targeted preprocessing of the images and signals effectively removes noise and outliers, enhances the accuracy of feature extraction, and improves the data quality; the early fusion strategy combines advanced algorithms to fuse multi-modal data at the feature level, fully excavating the correlations between data and enriching the feature vector information; the attention mechanism is introduced to automatically focus on sensitive areas, combined with adaptive adjustment, improving the pertinence and adaptability of feature extraction, and enhancing the robustness of the model to different working conditions; a dynamic prediction model is constructed, and through rate-of-change calculation and threshold setting, potential anomalies are accurately identified and reports are output in a timely manner to ensure the assembly quality of the gear pump.

[0017] 3. For this hydraulic gear pump assembly detection system, images are collected at low speed and information is synchronously recorded. After grouping by angle, a precise 3D model of the gear pump is constructed using advanced 3D reconstruction technology, providing a reliable basis for subsequent comparative analysis; the clearance data of the 3D model is compared with the predicted data, the deviation is calculated and visually displayed, intuitively presenting the assembly error and clearance distribution, facilitating rapid problem positioning; the mean and standard deviation of the gear clearance are calculated to determine the fluctuation value, identifying periodic characteristics, which helps to deeply understand the clearance change of the gear pump during operation; the deviation is calculated angle by angle, a threshold is set to mark the abnormal angle, and the error source is analyzed in combination with the jitter characteristics, which can accurately find the root cause of the assembly problem and provide a strong basis for improving the assembly quality.

[0018] 4. The hydraulic gear pump assembly detection system collects data at different speeds and temperatures, analyzes its effects on clearance fluctuations, jitter frequencies, and thermal expansion, establishes a correlation model to comprehensively understand the performance of the gear pump; modifies the prediction model parameters based on temperature effects to improve the accuracy of the model under different working conditions, enhancing the adaptability and reliability of the model; constructs a digital twin model to map actual data in real time, simulate the operating states under different working conditions, predict performance and reliability, and compare and verify with the actual situation to provide strong support for optimizing the assembly process; introduces a reinforcement learning algorithm to intelligently adjust assembly parameters, combines multi-dimensional data to locate error sources, outputs optimization suggestions, and effectively improves the quality and reliability of the gear pump. Description of the Drawings

[0019] Figure 1 It is a system module block diagram of the present invention; Figure 2 It is a schematic diagram of the working process of the prediction model establishment module of the present invention; Figure 3 It is a schematic diagram of the working process of the actual transmission comparison module of the present invention; Figure 4 It is a schematic diagram of the working process of the multi-condition extended analysis module of the present invention. Detailed Embodiments

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0021] The present invention discloses a hydraulic gear pump assembly detection system and provides the following four technical solutions: Figure 1 The first implementation manner is shown as follows: including: Prediction model establishment module: Collect multi-modal data of images, vibrations, and pressures and perform preprocessing; adopt an early fusion strategy to fuse images, vibration signals, and pressure data at the feature layer; adaptively extract gear tooth profile and clearance width features, and introduce an attention mechanism to automatically focus on the area most sensitive to clearance changes; construct a mathematical model to predict clearance changes, and perform model training and optimization; calculate the change rate of the predicted clearance, identify potential anomalies, and output a predicted clearance data set or an anomaly report; Actual transmission comparison module: Drive the gear pump to run at low speed, synchronously record images, timestamps, and angle information, and use 3D reconstruction technology to construct a 3D model of the gear pump; compare the predicted data with the clearance data in the 3D model; calculate the mean and standard deviation of the gear clearance, determine the fluctuation value, and identify periodic characteristics; compare the actual clearance data with the predicted value, calculate the deviation, mark abnormal angles, and analyze the error sources. Multi-condition extended analysis module: Collect data at different speeds, analyze the influence of speed on clearance fluctuation and jitter frequency; monitor the working temperature of the gear pump, analyze the influence of thermal expansion on the clearance, and correct the parameters of the prediction model; construct a digital twin model of the gear pump to simulate the operating states under different conditions; introduce a reinforcement learning algorithm to optimize the assembly process parameters; combine multi-dimensional data to locate the main error sources and output assembly optimization suggestions.

[0022] Through early fusion of multi-modal data and adaptive feature extraction, combined with the attention mechanism to focus on sensitive areas, a precise prediction model is constructed, which can effectively identify potential anomalies in advance, output the prediction data set or anomaly report in advance, and improve the detection efficiency; while the actual transmission comparison module drives the gear pump to run at low speed, constructs a model using 3D reconstruction technology, further compares the predicted and actual clearance data, calculates the deviation, marks abnormal angles, and analyzes the error sources, making the problem positioning more accurate; finally, the multi-condition extended analysis module considers the influence of speed and temperature, corrects the model parameters, constructs a digital twin model to simulate different conditions, introduces a reinforcement learning algorithm to optimize the assembly process parameters, combines multi-dimensional data to locate the error sources and outputs optimization suggestions, improving the assembly efficiency; and conducts dynamic detection and combines multi-parameter analysis for more accurate anomaly analysis.

[0023] Figure 2 The second implementation mode is shown, and the main difference from the first implementation mode is that the work content of the prediction model establishment module specifically includes: Multi-modal data acquisition and fusion: Data acquisition: Use a high-resolution industrial camera (resolution such as 2048×1536) to collect images of the assembled gear pump, covering the full angle of 0° to 360°, and define the angle step B, and collect a total of images; Integrate a vibration sensor (sampling frequency set to 1000Hz) and a pressure sensor (accuracy of ±0.1%), and synchronously collect the vibration signal V(t) and pressure data P(t) during the operation of the gear pump; The high-resolution industrial camera collects images at full angle, combined with the synchronous data collection of the vibration and pressure sensors, providing a rich and accurate information source for subsequent analysis.

[0024] Data preprocessing: Preprocess the image by denoising (using the Gaussian filtering algorithm with standard deviation σ = 1), grayscale conversion, and edge enhancement (using Canny edge detection with thresholds T1 = 50 and T2 = 150), and extract the gear profile and clearance features; Filter the vibration signal and pressure data (using low-pass filtering with cut-off frequency f c = 50 Hz) and perform normalization to remove noise and outliers; Perform targeted preprocessing on the image and signal to effectively remove noise and outliers, enhance the accuracy of feature extraction, and improve data quality.

[0025] Multi-modal fusion: Adopt the early fusion strategy and use a multi-modal fusion algorithm (such as the combination of convolutional neural network and recurrent neural network) to fuse the preprocessed image data I, vibration signal V, and pressure data P at the feature level. The fused feature vector F is expressed as: F = Fusion(I, V, P) = W I ·f I (I) + W V ·f V (V) + W P ·f P (P); Among them, f I 、f V 、f P are the feature extraction functions for the image, vibration signal, and pressure data respectively, and W I 、W V 、W P are the weight coefficients.

[0026] The early fusion strategy combines advanced algorithms to fuse multi-modal data at the feature level, fully mining the associations between data and enriching the information of the feature vector.

[0027] Adaptive feature extraction: Feature extraction: Extract the key parameters of the gear tooth profile and clearance width based on deep learning (such as YOLO, Mask R-CNN) or traditional CV algorithms (such as Canny edge detection + Hough transform); Introduce the attention mechanism to automatically focus on the regions in the image that are most sensitive to clearance changes. The attention weight A is calculated as: A = Softmax(W a ·h); Among them, h is the feature vector, W a is the weight matrix of the attention mechanism, and the Softmax function is used to calculate the attention weight A to ensure that the sum of all weights is 1, thereby representing the relative probabilities of different regions' sensitivities to clearance changes; Adaptive adjustment: According to the different working states and image characteristics of the gear pump, automatically adjust the parameters and strategies of feature extraction. Use the adaptive learning rate adjustment algorithm (such as the Adam optimizer), and the learning rate η is dynamically adjusted according to the gradient magnitude: ; where η0 is the initial learning rate, β1 and β2 are momentum parameters, m t , v t are the first-order and second-order momentum respectively, is a small constant to prevent division by zero.

[0028] Introduce the attention mechanism to automatically focus on sensitive areas, combine with adaptive adjustment, improve the pertinence and adaptability of feature extraction, and enhance the robustness of the model to different working conditions.

[0029] Construct a mathematical model and identify potential anomalies: Model selection: Construct a dynamic transmission prediction model (such as LSTM or Transformer), with the input being the angle sequence θ = [θ1, θ2, …, θ N and the fused multi-modal feature F, and the output being the predicted clearance value corresponding to the angle ; Model training: Use historical data to train the prediction model and optimize the model parameters; the loss function L can be defined as the mean squared error: ; where g i is the actual clearance value; Rate of change calculation: Calculate the rate of change r i of the predicted clearance with respect to the angle: ; Set the rate of change mutation threshold T r , if the rate of change mutation ∣r i ∣ > T r , then it is determined that there is a potential anomaly; Anomaly prediction: If there is no anomaly, output the predicted clearance data set; if there is an anomaly, mark the abnormal angle and output an anomaly report.

[0030] Construct a dynamic prediction model, accurately identify potential anomalies through rate of change calculation and threshold setting, output reports in a timely manner, and ensure the assembly quality of the gear pump.

[0031] Figure 3 The third implementation mode is shown, and the main difference from the first implementation mode is that the specific work content of the actual transmission comparison module includes: Collect images at low speed and construct a three-dimensional model: Drive and Acquisition: Operate the gear pump at a low speed (e.g., 10 rpm), set the number of frames to be acquired (e.g., 10 frames per second), and synchronously record the timestamp t i and the angle information θ i ; Image Grouping: Group the images by angle and extract the set of images I with the same angle θ ; 3D Reconstruction: Use 3D reconstruction techniques (such as structured light scanning or stereo vision) to reconstruct the acquired 2D images into a 3D model M of the gear pump

[0032] Collect images at a low speed and synchronously record information. After grouping by angle, use advanced 3D reconstruction techniques to construct an accurate 3D model of the gear pump, providing a reliable basis for subsequent comparative analysis

[0033] Comparative Analysis Based on the 3D Model and Actual Data Comparative Model Analysis: Compare the clearance data g in the 3D model M and the predicted data , and calculate the deviation e i : ; Through visualization tools, visually display the assembly errors and clearance distributions Compare the clearance data of the 3D model with the predicted data, calculate the deviation and visually display it, intuitively presenting the assembly errors and clearance distributions for easy quick problem location

[0034] Clearance Fluctuation Calculation: For the set of images I at the same angle θ , calculate the mean μ of the gear clearance g and the standard deviation σ g : ; ; Combine the mean μ g , the standard deviation σ g , engineering experience, and design requirements to determine the fluctuation value (e.g., ±0.05 mm); Periodic Feature Recognition Through Fourier transform or wavelet analysis, identify the periodic feature f of the clearance fluctuation p , and correlate with the gear jitter situation Calculating the mean and standard deviation of the gear clearance to determine the fluctuation value and identifying the periodic feature helps to deeply understand the clearance change situation during the operation of the gear pump

[0035] Deviation Calculation for Each Angle: Compare the actual clearance data g with the predicted value angle by angle, and calculate the deviation e i(if the absolute error ≤ 0.03mm is qualified); Abnormal marking and analysis: Set the deviation threshold. If the deviation exceeds the threshold, mark it as an abnormal angle, and analyze the error source in combination with the jitter characteristics (such as assembly eccentricity, tooth profile wear).

[0036] Calculating the deviation angle by angle, setting the threshold to mark the abnormal angle, and analyzing the error source in combination with the jitter characteristics can accurately find the root cause of the assembly problem and provide a strong basis for improving the assembly quality.

[0037] Figure 4 The fourth implementation mode is shown, and the main difference from the first implementation mode is that the working content of the multi-condition expansion analysis module specifically includes: Analysis of speed sensitivity: Data acquisition: Repeat the acquisition and analysis process under low-speed transmission, and continue to acquire data at medium speed (50rpm) and high speed (100rpm); Influence analysis: Analyze the influence of speed v on the standard deviation of clearance fluctuation σ g and jitter frequency f d , and establish a speed-error correlation model: σ g =a·v + b; f d =c·v d ; where a, b, c, d are model parameters.

[0038] By collecting data at different speeds and temperatures, analyzing its influence on clearance fluctuation, jitter frequency and thermal expansion, and establishing a correlation model, the understanding of the performance of the gear pump can be more comprehensive.

[0039] Analysis of temperature sensitivity: where a, b, c, d are model parameters.

[0040] By collecting data at different speeds and temperatures, analyzing its influence on clearance fluctuation, jitter frequency and thermal expansion, and establishing a correlation model, the understanding of the performance of the gear pump can be more comprehensive.

[0041] Analysis of temperature sensitivity: Data comparison: Repeat image acquisition and comparison at different temperatures, and analyze the influence of thermal expansion on the clearance g; the coefficient of thermal expansion α can be expressed as: ; where Δg is the change in clearance, ΔT is the change in temperature, and g0 is the initial clearance; Model correction: According to the temperature influence, correct the prediction model parameters to improve the accuracy of the model at different temperatures.

[0042] Modify the parameters of the prediction model according to the influence of temperature to improve the accuracy of the model under different working conditions and enhance the adaptability and reliability of the model.

[0043] Build a digital twin model: Model construction: Build a digital twin model M' of the gear pump and map the actually collected data to the digital twin model in real time; Simulation and verification: Predict the performance and reliability by simulating the operating state of the gear pump under different working conditions; and compare and verify with the actual test results to optimize the assembly process.

[0044] Build a digital twin model, map actual data in real time, simulate the operating state under different working conditions, predict performance and reliability, and compare with the actual situation for verification, providing strong support for optimizing the assembly process.

[0045] Perform assembly optimization based on reinforcement learning: Agent definition: Introduce a reinforcement learning algorithm, take the assembly process parameters p (such as bearing preload, tooth profile machining accuracy) as the actions of the agent, and take the test result r as the reward signal; Optimization process: Continuously try different combinations of assembly parameters to let the agent learn the assembly process; set the reward function R. The test result r includes the deviation of the clearance and the jitter frequency, and the calculation formula of R is: ; where λ is the weight coefficient to balance the influence of the clearance deviation and the jitter frequency; Gradually optimize the assembly parameters: Use the policy gradient algorithm (such as REINFORCE) to update the assembly parameters to improve the quality and reliability of the gear pump; Data analysis: Combine the three-dimensional data of angle, speed, and temperature, and locate the main error sources through PCA or decision tree algorithms; Optimization suggestions: Output assembly optimization suggestions (such as adjusting the bearing preload, improving the tooth profile machining accuracy) in combination with historical experience.

[0046] Introduce a reinforcement learning algorithm, intelligently adjust the assembly parameters, locate the error sources in combination with multi-dimensional data, and output optimization suggestions to effectively improve the quality and reliability of the gear pump.

[0047] A certain factory originally used traditional manual inspection methods to detect the clearance and evaluate the assembly quality of hydraulic gear pumps, which had problems such as low detection efficiency, large errors, and difficulty in comprehensively covering all working conditions. To improve the detection efficiency and accuracy, the factory decided to adopt this solution for the inspection of hydraulic gear pumps. The comparison of the inspection results with the original process is shown in Table 1 below: Table 1 Comparison table of inspection results

[0048] In summary, compared with the original method, when a certain factory uses this solution to inspect hydraulic gear pumps, it has significantly improved detection efficiency, smaller detection errors, a more comprehensive operating condition coverage rate, a higher abnormal recognition accuracy rate, more effective assembly optimization suggestions, and lower detection costs.

[0049] At the same time, the content not described in detail in this specification belongs to the prior art well-known to those skilled in the art.

[0050] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0051] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A hydraulic gear pump assembly detection system, characterized in that, Including: Prediction model establishment module: Collect multi-modal data of images, vibrations, and pressures and perform preprocessing; Adopt an early fusion strategy to fuse images, vibration signals, and pressure data at the feature level; Adaptively extract gear tooth profile and clearance width features, and introduce an attention mechanism to automatically focus on the area most sensitive to clearance changes; Construct a mathematical model to predict clearance changes, and perform model training and optimization; Calculate the change rate of the predicted clearance, identify potential anomalies, and output a predicted clearance dataset or an anomaly report; Actual transmission comparison module: Drive the gear pump to run at a low speed, synchronously record images, timestamps, and angle information, and use 3D reconstruction technology to construct a 3D model of the gear pump; Compare the predicted data with the clearance data in the 3D model; Calculate the mean and standard deviation of the gear clearance, determine the fluctuation value, and identify periodic features; Compare the actual clearance data with the predicted value, calculate the deviation, mark the abnormal angle, and analyze the error source; Multi-condition extended analysis module: Collect data at different speeds, analyze the influence of speed on clearance fluctuations and jitter frequencies; Monitor the working temperature of the gear pump, analyze the influence of thermal expansion on the clearance, and correct the prediction model parameters; Construct a digital twin model of the gear pump to simulate the operating state under different conditions; Introduce a reinforcement learning algorithm to optimize the assembly process parameters; Combine multi-dimensional data to locate the main error sources and output assembly optimization suggestions.

2. The assembly detection system for a hydraulic gear pump according to claim 1, characterized in that: The content of multi-modal data collection and fusion performed by the prediction model establishment module includes: Data collection: Use a high-resolution industrial camera to collect images of the assembled gear pump, covering the full angle from 0° to 360°, and define the angular step B, and a total of images are collected; Integrate vibration sensors and pressure sensors to synchronously collect vibration signal V(t) and pressure data P(t) during the operation of the gear pump; Data preprocessing: Perform preprocessing steps of denoising, grayscaling, and edge enhancement on the image to extract gear contour and clearance features; Perform filtering and normalization processing on the vibration signal and pressure data to remove noise and outliers; Multi-modal fusion: Adopt an early fusion strategy, use a multi-modal fusion algorithm, and fuse the preprocessed image data I, vibration signal V, and pressure data P at the feature level. The fused feature vector F is expressed as: F = Fusion(I, V, P) = W I ·f I (I) + W V ·f V (V) + W P ·f P (P); Among them, f I , f V , f P are the feature extraction functions for images, vibration signals, and pressure data respectively. W I , W V , W P are weight coefficients.

3. The assembly detection system of a hydraulic gear pump according to claim 2, characterized in that: The content of adaptive feature extraction performed by the prediction model establishment module includes: Feature extraction: Extract key parameters of gear tooth profile and clearance width based on deep learning or traditional CV algorithms; A = Softmax(W a ·h); where h is the feature vector, and W a is the weight matrix of the attention mechanism. The Softmax function is used to calculate the attention weight A, ensuring that the sum of all weights is 1, thereby representing the relative probability of the sensitivity of different regions to the gap change; Introduce an attention mechanism to automatically focus on the area in the image that is most sensitive to clearance changes. The attention weight A is calculated as: ； Among them, η0 is the initial learning rate, β1 and β2 are momentum parameters, m t , v t are the first-order and second-order momenta respectively, is a small constant to prevent division by zero.

4. The assembly detection system for a hydraulic gear pump according to claim 1, wherein: Adaptive adjustment: Automatically adjust the parameters and strategies of feature extraction according to the different working states and image characteristics of the gear pump, and use an adaptive learning rate adjustment algorithm. The learning rate η is dynamically adjusted according to the gradient magnitude: Model selection: Build a dynamic transmission prediction model. The inputs are the angle sequence θ = [θ1, θ2, …, θ N and the fused multi-modal feature F, and the output is the clearance prediction value corresponding to the angle ; The content of constructing a mathematical model and identifying potential anomalies performed by the prediction model establishment module includes: ； Among them, g i is the actual clearance value; Rate of change calculation: Calculate the rate of change r of the predicted gap with respect to the angle i : ； Set the mutation threshold T for the change rate r , if the change rate mutation |r i | > T r , it is determined that there is a potential anomaly; Model training: Use historical data to train the prediction model and optimize the model parameters; The loss function L can be defined as the mean squared error:

5. A hydraulic gear pump assembly detection system according to claim 1, characterized in that: Anomaly prediction: If there is no anomaly, output a predicted clearance dataset; If there is an anomaly, mark the abnormal angle and output an anomaly report. Drive and acquisition: Drive the gear pump to run at a low speed, set the number of frames to be acquired, and synchronously record the timestamp t i and the angle information θ i ; Image grouping: Group by angle and extract the set of images I with the same angle θ ; The content of image collection and 3D model construction by the actual transmission comparison module at low speed includes: 3D reconstruction: Use 3D reconstruction technology to reconstruct the collected 2D image into a 3D model M of the gear pump.

6. The assembly detection system of a hydraulic gear pump according to claim 5, characterized in that: The content of the actual transmission comparison module for comparative analysis based on the 3D model and actual data includes: Comparison model analysis: Compare the gap data g in the three-dimensional model M with the predicted data to calculate the deviation e i : ； Through visualization tools, intuitively display the assembly error and clearance distribution; Clearance fluctuation calculation: For the image set I at the same angle θ , calculate the mean value μ of the gear clearance g and the standard deviation σ g : ； ； Combined with the mean value μ g , the standard deviation σ g , engineering experience and design requirements to determine the fluctuation value; Periodic feature recognition: Identify the periodic characteristics f of the gap fluctuation through Fourier transform or wavelet analysis p , and correlate the gear jitter situation; Calculation of angular deviation: Compare the actual clearance data g with the predicted value angle by angle, and calculate the deviation e i ; Abnormal marking and analysis: Set the deviation threshold. If the deviation exceeds the threshold, mark it as an abnormal angle, and analyze the error source in combination with the jitter characteristics.

7. A hydraulic gear pump assembly detection system according to claim 1, characterized in that: The content of the multi-condition extended analysis module for the analysis of speed sensitivity includes: Data acquisition: Repeat the acquisition and analysis process under low-speed transmission, and continue to acquire data at medium speed and high speed; Influence analysis: Analyze the influence of the speed v on the standard deviation σ of the gap fluctuation g and the jitter frequency f d , and establish a speed-error correlation model: σ g = a·v + b; f d = c·v d ; Among them, a, b, c, and d are model parameters.

8. The assembly detection system of a hydraulic gear pump according to claim 7, characterized in that: The content of the multi-condition extended analysis module for the analysis of temperature sensitivity includes: Temperature monitoring: Integrate a temperature sensor to monitor the working temperature T of the gear pump; Data comparison: Repeat image acquisition and comparison at different temperatures, and analyze the influence of thermal expansion on the clearance g; The coefficient of thermal expansion α can be expressed as: ； Among them, Δg is the change in clearance, ΔT is the change in temperature, and g0 is the initial clearance; Model correction: According to the temperature influence, correct the prediction model parameters to improve the accuracy of the model at different temperatures.

9. The assembly inspection system for a hydraulic gear pump according to claim 8, wherein: The content of the multi-condition extended analysis module for constructing a digital twin model includes: Model construction: Construct a digital twin model M' of the gear pump, and map the actually collected data to the digital twin model in real time; Simulation and verification: Predict the performance and reliability by simulating the operating state of the gear pump under different conditions; and compare and verify with the actual detection results to optimize the assembly process.

10. A hydraulic gear pump assembly detection system according to claim 9, characterized in that: The content of the multi-condition extended analysis module for assembly optimization based on reinforcement learning includes: Agent definition: Introduce a reinforcement learning algorithm, take the assembly process parameter p as the action of the agent, and take the detection result r as the reward signal; Optimization process: Continuously try different combinations of assembly parameters to let the agent learn the assembly process; Set the reward function R. The detection result r includes the deviation and jitter frequency of the clearance, then the calculation formula of R is: ； Among them, λ is the weight coefficient to balance the influence of clearance deviation and jitter frequency; Gradually optimize the assembly parameters: Use the policy gradient algorithm to update the assembly parameters to improve the quality and reliability of the gear pump; Data analysis: Combine the three-dimensional data of angle, speed, and temperature, and locate the main error source through PCA or decision tree algorithm; Optimization suggestions: Output assembly optimization suggestions in combination with historical experience.

Images