Finish machining treatment system and method for gear production and manufacturing

Abstract

The invention relates to the technical field of gear finish machining, in particular to a gear production and manufacturing finish machining treatment system and method.The method comprises the steps that historical machining data of a high-precision transmission gear is acquired through a historical data acquisition module; and a stage surface image after the current processing stage on the first production line is completed is obtained through a real-time data acquisition module. And the stage detection module generates stage real-time detection data according to the stage surface image, and obtains a stage detection result in combination with standard detection data. The classification module classifies the gears according to the detection result, removes waste products and transfers process adjustment products and rework products to a second production line. The processing parameter adjusting module carries out clustering according to the real-time detection data of the process adjustment product and the rework product, and generates processing adjustment parameters and rework adjustment parameters. The second production line control module adjusts the second production line according to the machining adjustment parameter and the reworking adjustment parameter. According to the invention, the production process is optimized, and the quality control and production efficiency of the gear are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gear finishing, and in particular to a gear production and manufacturing finishing processing system and method. Background Art

[0002] Gears are one of the core components in mechanical transmission systems. Although gears are small, their manufacturing quality directly affects the operating efficiency and reliability of the entire mechanical equipment where the gears are located. In the production process of gears, it is usually necessary to go through multiple complex processing steps, including processing, heat treatment and precision grinding of gear blanks. Each step puts forward strict requirements on the dimensional accuracy, surface quality and material properties of the gears. High-precision transmission gears are a type of gear, which are widely used in the fields of automobiles, aviation and precision machinery. High-precision transmission gears have extremely high requirements for manufacturing accuracy and product stability. Among them, heat treatment (such as quenching, tempering, etc.) often changes the hardness and properties of the gear material, which may cause material deformation or dimensional deviation. Therefore, high-precision transmission gears need to be fine-machined after heat treatment to correct the deformation caused by the heat treatment process and ensure that the dimensional accuracy, surface quality and tooth shape of the gears meet the design requirements. However, in the actual production process, the production line is often affected by random factors such as material properties, equipment performance and operating conditions, resulting in fluctuations in process parameters at certain stages, which in turn affects the final quality of the gears.

[0003] Many methods for precision defect detection of gears have been proposed in the prior art, which can timely identify quality problems such as surface flaws and dimensional deviations of gears during the processing. However, the prior art mainly focuses on detecting quality problems of gears, but often lacks effective follow-up measures for handling abnormal gears detected.

[0004] Specifically, the existing technology has made certain progress in the precision detection and process control of gear production, and many systems can achieve high-precision defect detection and monitoring of process parameters. However, these technologies still have certain limitations in practical applications. For example, the efficiency of real-time feedback and processing needs to be improved. Although the detection system can identify anomalies, it is still insufficient in how to quickly respond to and handle these anomalies. In addition, although there are some means of intelligent diversion and graded processing in the existing technology, it is still not perfect in terms of flexible adjustment for defects of different degrees. At the same time, the dynamic adjustment and optimization mechanism of process parameters is also relatively limited, and it cannot fully adapt to the random fluctuations in the production process. Therefore, although the existing technology has basic functions, there is still room for improvement in further improving the requirements of flexibility, response speed and adaptability, resulting in the failure to timely handle or feedback the occasional defective gears in the production process, affecting the overall product reliability and stability of product quality.

[0005] To this end, a gear production and manufacturing finishing processing system and method are proposed. Summary of the invention

[0006] The object of the present invention is to provide a gear production and finishing processing system and method, the system includes multiple modules for optimizing the detection and adjustment in the production process of high-precision transmission gears. The historical data acquisition module obtains the historical processing data of high-precision transmission gears; the real-time data acquisition module obtains the surface image after each processing stage in real time and generates the stage real-time detection data; the processing stage includes the raw material preparation stage, the blank processing stage, the heat treatment stage and the finishing stage; the stage detection module analyzes the stage real-time detection data and compares it with the corresponding standard detection data to obtain the stage detection results; the classification module divides the gears into qualified products, process adjustment products, rework products and waste products according to the detection results, and moves the adjusted products and rework products to the second production line. The processing parameter adjustment module clusters the real-time detection data of the process adjustment products and rework products in each processing stage, and obtains multiple processing batches and multiple rework batches respectively; for each processing batch, the processing adjustment parameters are obtained according to the corresponding processing stage and the stage detection results; for each rework batch, the rework adjustment parameters are determined according to the corresponding processing stage and the stage detection results; the second production line control module adjusts the second production line accordingly according to the processing adjustment parameters and the rework adjustment parameters. At the same time, the present invention proposes a stage classification model to accurately classify the specific process requirements and test results of different stages, realize more detailed and flexible classification processing, and improve the overall quality and efficiency of gear production. By preprocessing and accurately detecting the surface images of the four stages of raw material preparation, blank processing, heat treatment and finishing, differentiated detection methods are adopted to meet the specific process requirements of each stage, ensure the quality monitoring of the whole process, maximize the capture of key quality parameters in each processing step, and improve the detection accuracy and production consistency. The present invention also introduces a multimodal fault prediction model based on sound, vibration and image, comprehensively evaluates the gear status through the finishing multimodal detection step, and combines the LightGBM model for fault prediction, further improving the detection accuracy and final factory quality of the product.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A gear production and finishing processing system, comprising:

[0009] Historical data acquisition module, used to obtain historical processing data of high-precision transmission gears;

[0010] A real-time data acquisition module is used to obtain in real time the high-precision transmission gear surface image after the current processing stage is completed on the first production line to obtain a stage surface image; the processing stage includes a raw material preparation stage, a blank processing stage, a heat treatment stage and a finishing stage;

[0011] A stage detection module, used to obtain stage real-time detection data according to the stage surface image; and obtain stage detection results according to the stage real-time detection data and corresponding standard detection data;

[0012] A classification module is used to classify the high-precision transmission gears according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; remove the waste products from the first production line; and move the process-adjusted products and the reworked products to the second production line;

[0013] A processing parameter adjustment module, for clustering the process-adjusted products and the reworked products according to the real-time detection data of the stages in each processing stage, respectively obtaining a plurality of processing batches and a plurality of rework batches; for each processing batch, obtaining a processing adjustment parameter according to the corresponding processing stage and the corresponding stage detection result; for each rework batch, determining a rework adjustment parameter according to the corresponding processing stage and the corresponding stage detection result;

[0014] The second production line control module is used to adjust the second production line according to the processing adjustment parameters and the rework adjustment parameters respectively.

[0015] Furthermore, the historical processing data of the high-precision transmission gear is obtained specifically as follows:

[0016] The historical processing data includes the operating parameters of the high-precision transmission gear at each processing stage, and the detection data, detection results and post-processing data after each processing stage is completed; the detection results include qualified and unqualified; the post-processing data includes subsequent processing stages and subsequent processing parameters;

[0017] Acquire the test data with qualified test results to obtain historical qualified data; analyze and extract key quality parameters based on the historical qualified data to generate the standard test data; the standard test data includes a set of key quality parameter ranges corresponding to each processing stage; the post-processing data of the historical qualified data is empty;

[0018] The test data with unqualified test results is obtained to obtain historical unqualified data; the post-processing data of the historical unqualified data includes a subsequent stage and subsequent stage processing adjustment parameters.

[0019] Furthermore, the high-precision transmission gears are classified according to the stage detection results as follows:

[0020] The stage detection results are input into the trained corresponding stage classification model to obtain the stage classification results; the training process of the stage classification model includes the following steps:

[0021] The historical qualified data and the historical unqualified data are used as training data; the training data is preprocessed, including missing value filling, outlier removal and data normalization, to obtain preprocessed training data; according to the detection results and the post-processing data, a stage label is assigned to each data sample in the preprocessed training data; the stage labels include qualified product labels, process adjustment product labels, rework product labels and waste product labels; the stage labels are used to train the stage classification model; feature engineering is performed according to the preprocessed training data to construct multiple surface size ratio features to obtain an extended feature set; principal component analysis is used to select key features from the extended feature set to obtain a selected feature set; the selected feature set is used as an independent variable, the stage label is used as a dependent variable, and the stage classification model is trained using a gradient boosting tree to obtain N stage classification models, where N is the number of the processing stages.

[0022] Furthermore, the stage surface image and the stage real-time detection data are specifically:

[0023] The stage surface images include the front image of the raw material preparation stage, the back image of the raw material preparation stage, the front image of the blank processing stage, the back image of the blank processing stage, the front image of the heat treatment stage, the back image of the heat treatment stage, the front image of the finishing stage and the back image of the finishing stage; the stage real-time detection data includes the raw material preparation stage detection data, the blank processing stage detection data, the heat treatment stage detection data and the finishing stage detection data.

[0024] Furthermore, the real-time detection data of the stage is obtained according to the surface image of the stage:

[0025] For the front image of the raw material preparation stage and the back image of the raw material preparation stage, the following operations are performed:

[0026] Preprocessing the front image of the raw material preparation stage and the back image of the raw material preparation stage to remove noise and uneven illumination, and obtain a standard raw material preparation image; extracting surface features and surface defect information from the standard raw material preparation image to obtain the raw material preparation stage detection data; the raw material preparation stage detection data includes surface roughness, the number of surface defects and surface color uniformity;

[0027] For the front image of the blank processing stage and the back image of the blank processing stage, the following operations are performed:

[0028] Preprocessing the front image of the blank processing stage and the back image of the blank processing stage to obtain a standard blank processing image; extracting the geometric contour and surface features of the blank from the standard blank processing image to obtain the blank processing stage detection data, wherein the blank processing stage detection data includes the outer diameter, length, width and thickness of the blank;

[0029] For the front image at the heat treatment stage and the back image at the heat treatment stage, the following operations are performed:

[0030] Preprocessing the front image of the heat treatment stage and the back image of the heat treatment stage to obtain a standard heat treatment image; using an image analysis algorithm to extract surface features from the standard heat treatment image to obtain the heat treatment stage detection data, wherein the heat treatment stage detection data includes a surface hardness value, a surface color change value, a surface crack length, a surface deformation amount, and a surface roughness;

[0031] For the front image at the finishing stage and the back image at the finishing stage, the following operations are performed:

[0032] The front image of the finishing stage and the back image of the finishing stage are preprocessed, including denoising, graying and edge enhancement, to obtain a standard finishing image; an edge detection algorithm is used to extract the outline of the gear from the standard finishing image, identify the key dimensions, and obtain an image processing result; the key dimensions include the tooth top circle, tooth root circle, tooth thickness, tooth pitch and tooth profile angle; according to the image processing result, the dimensional parameters of the high-precision transmission gear are calculated; the dimensional parameters include tooth height, tooth pitch, tooth thickness, gear inner diameter and outer diameter; according to the resolution and imaging ratio of the camera that collects the surface image of the stage, the dimensional parameters are calibrated, the size of the pixel unit is converted into the actual physical size, and the calibrated dimensional parameters are obtained; based on the extracted key dimensions and the calibrated dimensional parameters, the actual physical dimensional parameters of the high-precision transmission gear are calculated to obtain the finishing stage detection data.

[0033] Furthermore, for each of the processing batches, obtaining the processing adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the processing batch respectively to obtain the batch processing quantity; if the batch processing quantity is greater than the corresponding stage production quantity threshold, determining the processing adjustment parameters of the processing batch according to the historical processing data.

[0034] Furthermore, for each of the rework batches, determining the rework adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the rework batch respectively to obtain the rework quantity; if the rework quantity is greater than the corresponding stage rework production quantity threshold, determining the rework adjustment parameters of the rework batch according to the historical processing data.

[0035] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step comprises:

[0036] Acquire the sound data, vibration data and surface image data of the high-precision transmission gear after the finishing stage, and acquire the historical sound data, historical vibration data, historical surface image data and annotation data of the high-precision transmission gear on the historical production line after the finishing stage; the surface image data and the historical surface image data both include the front image of the gear and the back image of the gear; the annotation data include the historical quality inspection results of the high-precision transmission gear, and the historical quality inspection results include faults and non-faults;

[0037] The historical sound data, the historical vibration data and the historical surface image data are processed using a training set feature engineering process, including: extracting features of the front image of the gear and the back image of the gear in the historical surface image data respectively by a convolutional neural network to obtain historical front features and historical back features; extracting frequency domain features of the historical sound data by a time-frequency analysis method; extracting time domain features of the historical sound data and the historical vibration data by a long short-term memory network, and obtaining historical sound features and historical vibration features after integration;

[0038] The extracted historical front features, historical back features, historical sound features, and historical vibration features are standardized to obtain historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features; the historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features are multimodally fused to generate historical fusion features;

[0039] According to the historical fusion features and the labeled data, the LightGBM model is used for supervised learning to train a fault prediction model to identify whether a gear is faulty.

[0040] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step further comprises:

[0041] Processing the sound data, the vibration data and the surface image data through the same steps as the training set feature engineering process, including: extracting front features and back features through the convolutional neural network, extracting frequency domain features of the sound data using the time-frequency analysis, extracting time domain features of the sound data and the vibration data using the long short-term memory network, and obtaining sound features and vibration features after integration;

[0042] The front feature, the back feature, the sound feature and the vibration feature are standardized using the same standardization processing method to obtain the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature; the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature are subjected to feature fusion in the same manner as the multimodal feature fusion to generate a fusion feature;

[0043] The fusion features are input into the trained fault prediction model to obtain prediction results; the prediction results include good gears and faulty gears.

[0044] A gear production and finishing method comprises the following steps:

[0045] Obtain historical processing data of high-precision transmission gears;

[0046] Real-time acquisition of high-precision transmission gear surface images after the current processing stage is completed on the first production line to obtain stage surface images; the processing stages include raw material preparation stage, blank processing stage, heat treatment stage and finishing stage;

[0047] Obtaining stage real-time detection data according to the stage surface image; obtaining stage detection results according to the stage real-time detection data and corresponding standard detection data;

[0048] Classify the high-precision transmission gears according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; remove the waste products from the first production line; move the process-adjusted products and the reworked products to the second production line;

[0049] Clustering is performed based on the real-time detection data of the process-adjusted products and the reworked products at each of the processing stages to obtain a plurality of processing batches and a plurality of rework batches respectively; for each of the processing batches, a processing adjustment parameter is obtained based on the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted based on the processing adjustment parameter; for each of the rework batches, a rework adjustment parameter is determined based on the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted based on the rework adjustment parameter.

[0050] Compared with the prior art, the present invention has the following beneficial effects:

[0051] 1. The present invention proposes a stage classification model. By training N stage classification models, it can accurately classify the specific process requirements and test results of each processing stage. The detection focus of gears in different processing stages is different, and there are also differences in surface features, dimensional parameters and quality standards. By training an independent classification model for each processing stage, the system can automatically identify qualified products, process adjustment products, rework products and scrap products according to the specific inspection data of each stage, and realize more detailed and flexible classification processing. This design can improve classification accuracy, optimize production processes, ensure timely adjustment and handling of abnormalities at each stage, and ultimately improve the overall quality and efficiency of gear production.

[0052] 2. The present invention ensures real-time monitoring and data feedback at each stage of the gear production process by preprocessing and accurately detecting the surface image at each processing stage. For the raw material preparation stage, blank processing stage, heat treatment stage and finishing stage. The detection data is acquired for each of the four different processing stages, and different acquisition methods are used to meet the specific process requirements of each stage, and to ensure comprehensive monitoring and precise control of the gear quality during the entire production process. The detection focus of each stage is different: the raw material preparation stage focuses on the surface quality and uniformity of the material, the blank processing stage focuses on monitoring the initial geometric dimensions and shapes, the heat treatment stage detects the surface hardness, deformation and cracks, and the finishing stage requires high-precision dimensional measurement and contour extraction. This staged and differentiated detection design can maximize the capture of key quality parameters in each processing step, avoid missing any potential problems, improve the detection accuracy and adaptability of the system, and ensure the quality consistency of the entire gear production process and the high reliability of the final product.

[0053] 3. The present invention proposes a multimodal vibration prediction model based on sound, vibration and image. Through the multimodal detection step of finishing, the system can conduct a comprehensive quality assessment of the gears in the final stage of the production process, thereby improving the reliability and detection accuracy of the product. Although surface image detection has been carried out in each processing stage, the sound data and vibration introduced in the finishing stage provide more detailed and multi-dimensional quality information, especially for minor defects and internal material stress that may be caused during the processing process, which further enriches the dimension and precision of the gear quality information. Through convolutional neural networks, time-frequency analysis and long short-term memory networks, data features of different modes are extracted, combined with multimodal feature fusion technology, the finishing status of the gear is comprehensively evaluated, and the LightGBM fault prediction model based on historical data training is used to accurately identify potential faults, effectively avoid quality problems, and ensure the high quality and consistency of the final gears shipped. The effective combination of comprehensive stage detection and precise final detection is achieved, the entire production process is optimized, and the overall quality control capability and stability of the product are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 A structural diagram of a gear production and finishing processing system provided by an embodiment of the present invention;

[0055] Figure 2 A flowchart of the processing phases of the first production line and the second production line provided by an embodiment of the present invention;

[0056] Figure 3 A flow chart of gear production quality control provided by an embodiment of the present invention;

[0057] Figure 4 A flow chart of a gear production and finishing method provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] High-precision transmission gears are gears used for efficient and precise power transmission. Their production requirements are more stringent than those of ordinary gears. Such gears are widely used in mechanical systems that require high reliability and performance, such as automobiles, aerospace, robotics, precision machinery, and industrial equipment. The manufacturing process of high-precision transmission gears has very strict control over dimensional tolerances, which can achieve higher precision. This high precision ensures accurate meshing of the gears during operation, reduces tooth clearance, and thus improves transmission efficiency.

[0060] Embodiment 1

[0061] Company A is a gear manufacturing plant that mainly produces high-precision transmission gears. It has several production lines, one of which uses a gear production and finishing system. Figure 1 As shown, the system includes the following modules:

[0062] Historical data acquisition module, used to obtain historical processing data of high-precision transmission gears;

[0063] The real-time data acquisition module is used to obtain the high-precision transmission gear surface image after the current processing stage is completed on the first production line in real time to obtain the stage surface image; the processing stage includes the raw material preparation stage, the blank processing stage, the heat treatment stage and the finishing stage; in this embodiment, Figure 2 As shown in the figure, the processing stage consists of raw material preparation stage, blank processing stage, heat treatment stage and finishing stage. The first production line is used for mass production of high-precision transmission gears. In order to ensure the stability and consistency of the production process, the operating parameters in the production process remain unchanged. At the same time, the historical processing data also comes from the same production line (the first production line), producing high-precision transmission gears of the same specification.

[0064] A stage detection module, used to obtain stage real-time detection data according to the stage surface image; and obtain stage detection results according to the stage real-time detection data and corresponding standard detection data;

[0065] A classification module is used to classify the high-precision transmission gears according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; Figure 3 As shown, the waste products are removed from the first production line; the process-adjusted products and the reworked products are moved to the second production line, wherein the process-adjusted products need to be processed further by adjusting the operating parameters of the subsequent processing stage, while the reworked products need to be reworked in the current processing stage; qualified products will directly enter the next processing stage. Figure 2, for the finishing stage, since it is the last stage in the production process, the high-precision transmission gears are classified according to the inspection results of the finishing stage to obtain qualified products, process-adjusted products, reworked products and scrapped products. Among them, the process-adjusted products are empty. The processing parameter adjustment module is used to cluster the real-time detection data of the process-adjusted products and the reworked products in each of the processing stages to obtain multiple processing batches and multiple rework batches respectively; for each of the processing batches, the processing adjustment parameters are obtained according to the corresponding processing stage and the corresponding stage inspection results; for each of the rework batches, the rework adjustment parameters are determined according to the corresponding processing stage and the corresponding stage inspection results;

[0066] The second production line control module is used to adjust the second production line according to the processing adjustment parameter and the rework adjustment parameter. Figure 2 After the defective products complete the stage processing on the second production line, they still need to undergo stage inspection and be classified again into qualified products, process-adjusted products, reworked products and waste products.

[0067] Table 1. Example table of historical processing data

[0068]

[0069] Furthermore, the historical processing data of the high-precision transmission gear is obtained specifically as follows:

[0070] Referring to Table 1, the historical processing data includes the operating parameters of the high-precision transmission gears in each processing stage, and the inspection data, inspection results and post-processing data after each processing stage is completed; the inspection results include qualified and unqualified; the post-processing data includes subsequent processing stages and subsequent processing parameters (performed on other production lines); It can be seen from Table 1 that the subsequent processing stage of the last recorded historical processing data is the same as the current completion stage, which is a rework product.

[0071] Table 2. Example of historical qualified data

[0072]

[0073] Table 3. Example table of historical non-conformance data

[0074]

[0075] The test data with qualified test results are obtained to obtain historical qualified data, as shown in Table 2; according to the historical qualified data, key quality parameters are analyzed and extracted to generate the standard test data; the standard test data includes a set of key quality parameter ranges corresponding to each processing stage; the post-processing data of the historical qualified data is empty;

[0076] The test data with unqualified test results are obtained to obtain historical unqualified data, as shown in Table 3; the post-processing data of the historical unqualified data includes a subsequent stage and subsequent stage processing adjustment parameters.

[0077] These historical processing data provide a solid foundation for quality control and process optimization in the production process. By recording the operating parameters, test data and post-processing information of each processing stage, the subsequent process flow of qualified and unqualified products can be identified and analyzed in a timely manner. Such data accumulation not only supports the generation of standard test data to ensure that the product quality at each stage meets the predetermined range, but also provides a basis for the rework and adjustment of unqualified products. For qualified products, extracting their key quality parameters to generate standard test data helps to provide a reliable quality benchmark for subsequent production. For unqualified products, by analyzing historical unqualified data and its subsequent processing stages and processing adjustment parameters, it can effectively guide the optimization of rework products on subsequent production lines to ensure that product quality is improved. This comprehensive data monitoring and analysis capability can significantly improve the efficiency of the production process and ultimately improve the overall quality and reliability of high-precision transmission gears.

[0078] Furthermore, the high-precision transmission gears are classified according to the stage detection results as follows:

[0079] The stage detection results are input into the trained corresponding stage classification model to obtain the stage classification results; the training process of the stage classification model includes the following steps:

[0080] The historical qualified data and the historical unqualified data of the current stage are used as training data; the training data are preprocessed, including missing value filling, outlier removal and data normalization, to obtain preprocessed training data; according to the detection results and the post-processing data, a stage label is assigned to each data sample in the preprocessing training data, as shown in Table 4; the stage labels include qualified product labels, process adjustment product labels, rework product labels and waste product labels; the stage labels are used to train the stage classification model; feature engineering is performed according to the preprocessing training data to construct multiple surface size ratio features to obtain an extended feature set; principal component analysis is used to select key features from the extended feature set to obtain a selected feature set; the selected feature set is used as an independent variable, the stage label is used as a dependent variable, and the stage classification model is trained using a gradient boosting tree to obtain N stage classification models, where N is the number of the processing stages.

[0081] In this embodiment, four stage classification models are obtained, namely, a raw material preparation stage classification model, a blank processing stage classification model, a heat treatment stage classification model and a finishing stage classification model.

[0082] Specifically, in the classification model of the raw material preparation stage, the constructed surface size ratio features include the ratio of length to width and the ratio of volume to thickness, etc.

[0083] In the classification model of the blank processing stage, the surface size ratio features constructed include the ratio of the blank thickness to the outer diameter and the ratio of the inner diameter to the outer diameter, etc.

[0084] In the heat treatment stage classification model, the surface dimension ratio features constructed include the ratio of tooth thickness to gear outer diameter, the ratio of tooth thickness to gear inner diameter, and the ratio of gear inner diameter to outer diameter;

[0085] In the classification model of the finishing stage, the surface dimension ratio features include: the ratio of tooth height to tooth pitch, which is used to evaluate the overall tooth shape accuracy of the gear; the ratio of tooth thickness to the inner diameter of the gear, which is used to evaluate the precision fit of the gear components; the ratio of tooth pitch to the outer diameter of the gear, which is used to determine whether the accuracy requirements of the gear in the transmission meet the standards.

[0086] In the classification model of the finishing stage, in addition to the surface dimension ratio features, adjacent dimension difference features are also constructed, including: adjacent tooth height difference, which is used to detect whether there is inconsistency in the gear tooth shape; adjacent tooth thickness difference, which is used to judge the thickness uniformity of each tooth part of the gear; adjacent tooth pitch difference, which is used to evaluate whether noise or jamming may occur during the gear transmission process.

[0087] Table 4. Training data example table

[0088]

[0089] Establishing and training a stage classification model can achieve accurate gear classification at each processing stage, ensuring quality control and efficiency improvement in the production process. By making full use of historical qualified and unqualified data, combined with the surface size ratio features constructed by feature engineering, the model can effectively capture the key dimensional changes and process characteristics of each processing stage. The principal component analysis method is used to extract key features, and the model is trained through the gradient boosting tree to ensure that the model has a strong classification ability and can accurately distinguish qualified products, process-adjusted products, reworked products and scrap products, providing a reliable basis for subsequent processing and effectively improving production efficiency and product quality consistency.

[0090] Furthermore, the stage surface image and the stage real-time detection data are specifically:

[0091] The stage surface images include the front image of the raw material preparation stage, the back image of the raw material preparation stage, the front image of the blank processing stage, the back image of the blank processing stage, the front image of the heat treatment stage, the back image of the heat treatment stage, the front image of the finishing stage and the back image of the finishing stage; the stage real-time detection data includes the raw material preparation stage detection data, the blank processing stage detection data, the heat treatment stage detection data and the finishing stage detection data.

[0092] By further refining the classification of stage surface images and real-time detection data, the system lays the foundation for subsequent stage detection, and is able to accurately detect and monitor gears at each key stage of the production process. The front and back images of each stage fully capture the subtle differences in the geometric features of the gear surface, while the stage real-time detection data provides real-time feedback on key size and shape changes. The production system can not only promptly identify defects that may occur during the processing process, but also quickly adjust production strategies based on the detection results of different stages, reducing unnecessary rework and scrap rates. Through multi-stage images and detection data, the quality control level of the entire production line can be improved, ensuring that the gears meet the predetermined standards at each processing stage, and improving the consistency and reliability of the final product.

[0093] Furthermore, the real-time detection data of the stage is obtained according to the surface image of the stage:

[0094] For the front image of the raw material preparation stage and the back image of the raw material preparation stage, the following operations are performed:

[0095] Preprocessing the front image of the raw material preparation stage and the back image of the raw material preparation stage to remove noise and uneven illumination, and obtain a standard raw material preparation image; extracting surface features and surface defect information from the standard raw material preparation image to obtain the raw material preparation stage detection data; the raw material preparation stage detection data includes surface roughness, the number of surface defects and surface color uniformity;

[0096] For the front image of the blank processing stage and the back image of the blank processing stage, the following operations are performed:

[0097] Preprocessing the front image of the blank processing stage and the back image of the blank processing stage to obtain a standard blank processing image; extracting the geometric contour and surface features of the blank from the standard blank processing image to obtain the blank processing stage detection data, wherein the blank processing stage detection data includes the outer diameter, length, width and thickness of the blank;

[0098] For the front image at the heat treatment stage and the back image at the heat treatment stage, the following operations are performed:

[0099] Preprocessing the front image of the heat treatment stage and the back image of the heat treatment stage to obtain a standard heat treatment image; using an image analysis algorithm to extract surface features from the standard heat treatment image to obtain the heat treatment stage detection data, wherein the heat treatment stage detection data includes a surface hardness value, a surface color change value, a surface crack length, a surface deformation amount, and a surface roughness;

[0100] For the front image at the finishing stage and the back image at the finishing stage, the following operations are performed:

[0101] The front image of the finishing stage and the back image of the finishing stage are preprocessed, including denoising, graying and edge enhancement, to obtain a standard finishing image; an edge detection algorithm is used to extract the outline of the gear from the standard finishing image, identify the key dimensions, and obtain an image processing result; the key dimensions include the tooth top circle, tooth root circle, tooth thickness, tooth pitch and tooth profile angle; according to the image processing result, the dimensional parameters of the high-precision transmission gear are calculated; the dimensional parameters include tooth height, tooth pitch, tooth thickness, gear inner diameter and outer diameter; according to the resolution and imaging ratio of the camera that collects the surface image of the stage, the dimensional parameters are calibrated, the size of the pixel unit is converted into the actual physical size, and the calibrated dimensional parameters are obtained; based on the extracted key dimensions and the calibrated dimensional parameters, the actual physical dimensional parameters of the high-precision transmission gear are calculated to obtain the finishing stage detection data.

[0102] By refining and processing the surface image in stages, combined with the corresponding detection data extraction and calibration, the present invention can achieve more accurate quality control and dimensional detection at each stage of gear production. For the raw material preparation stage, through image preprocessing and surface feature extraction, the system can effectively detect surface roughness, number of defects and color uniformity to ensure the quality of raw materials. In the blank processing stage, through the extraction of geometric contours and surface features, the system can accurately detect the outer diameter, length, width and thickness of the blank to ensure the basic dimensional accuracy of subsequent processing. In the heat treatment stage, the system uses image analysis algorithms to extract key parameters such as surface hardness, color change, crack length, deformation, etc., to promptly detect defects that may occur during the heat treatment process, and evaluate the quality based on surface roughness. In the finishing stage, the system extracts the key dimensions of the gear through the edge detection algorithm, calibrates it in combination with the camera's resolution and imaging ratio, accurately calculates the actual physical dimensions, and ensures that the finished gear meets the design specifications. This multi-stage, refined detection method not only improves the intelligence and automation level of the production process, greatly reduces the errors and delays of traditional detection, but also significantly improves the quality consistency of the product through accurate detection of key dimensions and surface features.

[0103] Furthermore, for each of the processing batches, obtaining the processing adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the processing batch respectively to obtain the batch processing quantity; if the batch processing quantity is greater than the corresponding stage production quantity threshold, determining the processing adjustment parameters of the processing batch according to the historical processing data.

[0104] When the batch processing quantity does not exceed the set production quantity threshold, the choice of suspending production and waiting for the quantity to accumulate before unified production can improve production efficiency and economic benefits. Through batch production, the time for frequent equipment startup and process switching can be reduced, the equipment utilization rate can be improved, and the production cost of a single product can be reduced. In addition, batch production can ensure that more products are manufactured under the same process conditions, reduce the errors caused by frequent process adjustments, and ensure the consistency of product quality. This strategy also optimizes production scheduling, avoids resource waste and unnecessary process switching caused by small batch production, and improves the stability and economic benefits of the production line as a whole.

[0105] Furthermore, for each of the rework batches, determining the rework adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the rework batch respectively to obtain the rework quantity; if the rework quantity is greater than the corresponding stage rework production quantity threshold, determining the rework adjustment parameters of the rework batch according to the historical processing data.

[0106] Similarly, this rework generation method can ensure that when the rework volume is large, the system can refer to historical data and dynamically adjust the process parameters to adapt to large-scale rework production and ensure the quality and consistency of the reworked products. At the same time, it reduces the frequent process adjustments caused by small-batch rework, improves the efficiency of the rework process, and reduces the rework cost. In addition, batch rework processing also reduces the frequent start and stop of equipment and the waste of resources, and improves the overall economic benefits and stability of the production line.

[0107] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step comprises:

[0108] Acquire the sound and vibration data of the gear in the test environment by establishing a simulated operating environment and using special test equipment. First, use the gear test bench to simulate the operation of the gear under actual working conditions, control the speed and load of the gear, and simulate its normal working state. Next, use a highly sensitive microphone to collect the sound data of the gear during operation, and ensure that the test environment is quiet to reduce noise interference. At the same time, install vibration sensors at key positions of the gear bearings or benches to collect vibration data of the gear during operation. Test under different speed and load conditions, record and save high-quality sound and vibration data. Finally, preprocess the collected data, such as removing noise and invalid signals, and use time-frequency analysis methods to extract the frequency domain features of the data. Through this process, the sound and vibration data of high-precision transmission gears under simulated operating conditions can be obtained for subsequent fault detection model training and product quality evaluation.

[0109] Acquire the sound data, vibration data and surface image data of the high-precision transmission gear after the finishing stage, and acquire the historical sound data, historical vibration data, historical surface image data and annotation data of the high-precision transmission gear on the historical production line after the finishing stage; the surface image data and the historical surface image data both include the front image of the gear and the back image of the gear; the annotation data include the historical quality inspection results of the high-precision transmission gear, and the historical quality inspection results include faults and non-faults;

[0110] The historical sound data, the historical vibration data and the historical surface image data are processed using a training set feature engineering process, including: extracting features of the front image of the gear and the back image of the gear in the historical surface image data respectively by a convolutional neural network to obtain historical front features and historical back features; extracting frequency domain features of the historical sound data by a time-frequency analysis method; extracting time domain features of the historical sound data and the historical vibration data by a long short-term memory network, and obtaining historical sound features and historical vibration features after integration;

[0111] The extracted historical front features, historical back features, historical sound features, and historical vibration features are standardized to obtain historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features; the historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features are multimodally fused to generate historical fusion features;

[0112] According to the historical fusion features and the labeled data, the LightGBM model is used for supervised learning to train a fault prediction model to identify whether a gear is faulty.

[0113] As the final product inspection before leaving the factory, by introducing the finishing multimodal inspection step, this system can fully integrate sound, vibration and surface image data to provide more accurate gear quality inspection and fault prediction. Through feature engineering to process the multimodal information in the historical data, including the extraction of image features by convolutional neural network, time-frequency analysis and the processing of sound and vibration data by long short-term memory network, the system can capture the full range of characteristics of the gear after finishing. After standardizing and fusing these features, a high-precision fault prediction model is trained by supervised learning using the LightGBM model. This multimodal detection method significantly improves the accuracy of fault identification, can effectively identify potential problems of gears, reduce missed detection and false detection rates, and ensure product quality consistency. At the same time, using historical data to train the model ensures that the prediction model has strong generalization ability and can adapt to different gear types and production environments, further optimizing the intelligence and automation level of the production line, and improving the overall production efficiency and quality management capabilities.

[0114] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step further comprises:

[0115] Processing the sound data, the vibration data and the surface image data through the same steps as the training set feature engineering process, including: extracting front features and back features through the convolutional neural network, extracting frequency domain features of the sound data using the time-frequency analysis, extracting time domain features of the sound data and the vibration data using the long short-term memory network, and obtaining sound features and vibration features after integration;

[0116] The front feature, the back feature, the sound feature and the vibration feature are standardized using the same standardization processing method to obtain the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature; the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature are subjected to feature fusion in the same manner as the multimodal feature fusion to generate a fusion feature;

[0117] The fusion features are input into the trained fault prediction model to obtain a prediction result; the prediction result includes good gears and faulty gears. Among them, the good gears can be shipped.

[0118] By introducing the finishing multimodal detection step, the system can use the same processing flow as the training set feature engineering to ensure consistency between real-time detection and historical data analysis. Through the convolutional neural network to extract the front and back image features, the time-frequency analysis to extract the frequency domain features of the sound data, and the long short-term memory network to process the time domain features of the sound and vibration data, the system can fully capture multimodal information and ensure the accuracy of data processing. Then, after standardization and feature fusion, the system inputs the fused features into the trained fault prediction model to accurately identify good gears and faulty gears. This process not only improves the accuracy of detection and fault prediction, but also significantly reduces the possibility of false detection and missed detection, ensuring stable product quality. In addition, good gears can be shipped directly after this inspection process, which simplifies the production process, improves production efficiency and the reliability of shipped products. Through the combination of multimodal fusion and high-precision prediction models, the system can provide real-time feedback on production conditions, optimize quality control processes, and greatly improve the intelligence level and overall benefits of the production line.

[0119] The gear manufacturing finishing processing system significantly improves production efficiency and product quality through automated and intelligent modular design. Using historical data acquisition modules and real-time data acquisition modules, the system can track and analyze the surface quality of gears at each processing stage in real time, detect problems in time, avoid defects accumulating in subsequent processes, and ensure that the processing quality meets the standards. By testing at each stage, the risk of rework and scrap can be reduced, while reducing the cost of correction. In addition, stage testing can improve production efficiency and prevent unqualified products from entering the next process. The classification module accurately classifies gear products and classifies high-precision transmission gears according to the test results of each processing stage to ensure that gears of different quality states are properly processed. The processing parameter adjustment module generates accurate processing and rework parameters through cluster analysis to ensure the processing accuracy of each production batch, and optimizes subsequent processing parameters through the adjustment of the second production line, continuously improves the production process, and improves production flexibility and efficiency. Overall, the system improves processing stability and reduces quality fluctuations.

[0120] Embodiment 2

[0121] Company B is an enterprise specializing in the manufacture of high-precision transmission gears, with multiple production lines. In order to improve production efficiency and product quality, Company B introduced an advanced gear production and finishing processing system on one of its production lines. The specific implementation process of the system is as follows:

[0122] A gear production and finishing method comprises the following steps:

[0123] See also Figure 4 S10 in the above example obtains historical processing data of high-precision transmission gears;

[0124] See also Figure 4 S20 in which a high-precision transmission gear surface image after the current processing stage on the first production line is completed is acquired in real time to obtain a stage surface image; the processing stage includes a raw material preparation stage, a blank processing stage, a heat treatment stage and a finishing stage;

[0125] See also Figure 4 In S30, obtaining stage real-time detection data according to the stage surface image; obtaining stage detection results according to the stage real-time detection data and corresponding standard detection data;

[0126] See also Figure 4 In S40, the high-precision transmission gears are classified according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; the waste products are removed from the first production line; and the process-adjusted products and the reworked products are moved to the second production line;

[0127] See also Figure 4 In S50, clustering is performed according to the real-time detection data of the process-adjusted products and the reworked products at each of the processing stages to obtain a plurality of processing batches and a plurality of rework batches respectively;

[0128] See also Figure 4 In S60, for each of the processing batches, a processing adjustment parameter is obtained according to the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted according to the processing adjustment parameter; for each of the rework batches, a rework adjustment parameter is determined according to the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted according to the rework adjustment parameter.

[0129] Furthermore, the historical processing data of the high-precision transmission gear is obtained specifically as follows:

[0130] The historical processing data includes the operating parameters of the high-precision transmission gear at each processing stage, and the detection data, detection results and post-processing data after each processing stage is completed; the detection results include qualified and unqualified; the post-processing data includes subsequent processing stages and subsequent processing parameters;

[0131] Acquire the test data with qualified test results to obtain historical qualified data; analyze and extract key quality parameters based on the historical qualified data to generate the standard test data; the standard test data includes a set of key quality parameter ranges corresponding to each processing stage; the post-processing data of the historical qualified data is empty;

[0132] The test data with unqualified test results is obtained to obtain historical unqualified data; the post-processing data of the historical unqualified data includes a subsequent stage and subsequent stage processing adjustment parameters.

[0133] Furthermore, the high-precision transmission gears are classified according to the stage detection results as follows:

[0134] The stage detection results are input into the trained corresponding stage classification model to obtain the stage classification results; the training process of the stage classification model includes the following steps:

[0135] The historical qualified data and the historical unqualified data are used as training data; the training data is preprocessed, including missing value filling, outlier removal and data normalization, to obtain preprocessed training data; according to the detection results and the post-processing data, a stage label is assigned to each data sample in the preprocessed training data; the stage labels include qualified product labels, process adjustment product labels, rework product labels and waste product labels; the stage labels are used to train the stage classification model; feature engineering is performed according to the preprocessed training data to construct multiple surface size ratio features to obtain an extended feature set; principal component analysis is used to select key features from the extended feature set to obtain a selected feature set; the selected feature set is used as an independent variable, the stage label is used as a dependent variable, and the stage classification model is trained using a gradient boosting tree to obtain N stage classification models, where N is the number of the processing stages.

[0136] Furthermore, the stage surface image and the stage real-time detection data are specifically:

[0137] The stage surface images include the front image of the raw material preparation stage, the back image of the raw material preparation stage, the front image of the blank processing stage, the back image of the blank processing stage, the front image of the heat treatment stage, the back image of the heat treatment stage, the front image of the finishing stage and the back image of the finishing stage; the stage real-time detection data includes the raw material preparation stage detection data, the blank processing stage detection data, the heat treatment stage detection data and the finishing stage detection data.

[0138] Furthermore, the real-time detection data of the stage is obtained according to the surface image of the stage:

[0139] For the front image of the raw material preparation stage and the back image of the raw material preparation stage, the following operations are performed:

[0140] Preprocessing the front image of the raw material preparation stage and the back image of the raw material preparation stage to remove noise and uneven illumination, and obtain a standard raw material preparation image; extracting surface features and surface defect information from the standard raw material preparation image to obtain the raw material preparation stage detection data; the raw material preparation stage detection data includes surface roughness, the number of surface defects and surface color uniformity;

[0141] For the front image of the blank processing stage and the back image of the blank processing stage, the following operations are performed:

[0142] Preprocessing the front image of the blank processing stage and the back image of the blank processing stage to obtain a standard blank processing image; extracting the geometric contour and surface features of the blank from the standard blank processing image to obtain the blank processing stage detection data, wherein the blank processing stage detection data includes the outer diameter, length, width and thickness of the blank;

[0143] For the front image at the heat treatment stage and the back image at the heat treatment stage, the following operations are performed:

[0144] Preprocessing the front image of the heat treatment stage and the back image of the heat treatment stage to obtain a standard heat treatment image; using an image analysis algorithm to extract surface features from the standard heat treatment image to obtain the heat treatment stage detection data, wherein the heat treatment stage detection data includes a surface hardness value, a surface color change value, a surface crack length, a surface deformation amount, and a surface roughness;

[0145] For the front image at the finishing stage and the back image at the finishing stage, the following operations are performed:

[0146] The front image of the finishing stage and the back image of the finishing stage are preprocessed, including denoising, graying and edge enhancement, to obtain a standard finishing image; an edge detection algorithm is used to extract the outline of the gear from the standard finishing image, identify the key dimensions, and obtain an image processing result; the key dimensions include the tooth top circle, tooth root circle, tooth thickness, tooth pitch and tooth profile angle; according to the image processing result, the dimensional parameters of the high-precision transmission gear are calculated; the dimensional parameters include tooth height, tooth pitch, tooth thickness, gear inner diameter and outer diameter; according to the resolution and imaging ratio of the camera that collects the surface image of the stage, the dimensional parameters are calibrated, the size of the pixel unit is converted into the actual physical size, and the calibrated dimensional parameters are obtained; based on the extracted key dimensions and the calibrated dimensional parameters, the actual physical dimensional parameters of the high-precision transmission gear are calculated to obtain the finishing stage detection data.

[0147] Furthermore, for each of the processing batches, obtaining the processing adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the processing batch respectively to obtain the batch processing quantity; if the batch processing quantity is greater than the corresponding stage production quantity threshold, determining the processing adjustment parameters of the processing batch according to the historical processing data.

[0148] Furthermore, for each of the rework batches, determining the rework adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the rework batch respectively to obtain the rework quantity; if the rework quantity is greater than the corresponding stage rework production quantity threshold, determining the rework adjustment parameters of the rework batch according to the historical processing data.

[0149] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step comprises:

[0150] Acquire the sound data, vibration data and surface image data of the high-precision transmission gear after the finishing stage, and acquire the historical sound data, historical vibration data, historical surface image data and annotation data of the high-precision transmission gear on the historical production line after the finishing stage; the surface image data and the historical surface image data both include the front image of the gear and the back image of the gear; the annotation data include the historical quality inspection results of the high-precision transmission gear, and the historical quality inspection results include faults and non-faults;

[0151] The historical sound data, the historical vibration data and the historical surface image data are processed using a training set feature engineering process, including: extracting features of the front image of the gear and the back image of the gear in the historical surface image data respectively by a convolutional neural network to obtain historical front features and historical back features; extracting frequency domain features of the historical sound data by a time-frequency analysis method; extracting time domain features of the historical sound data and the historical vibration data by a long short-term memory network, and obtaining historical sound features and historical vibration features after integration;

[0152] The extracted historical front features, historical back features, historical sound features, and historical vibration features are standardized to obtain historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features; the historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features are multimodally fused to generate historical fusion features;

[0153] According to the historical fusion features and the labeled data, the LightGBM model is used for supervised learning to train a fault prediction model to identify whether a gear is faulty.

[0154] Furthermore, the system further comprises a finishing multi-modal detection step, and the finishing multi-modal detection step further comprises:

[0155] The sound data, the vibration data and the surface image data are processed through the same steps as the training set feature engineering process, including: extracting front features and back features through the convolutional neural network, extracting frequency domain features of the sound data using the time-frequency analysis, extracting time domain features of the sound data and the vibration data using the long short-term memory network, and integrating them to obtain sound features and vibration features; in this embodiment, the sound data is collected by a high-precision microphone to record the sound signal when the high-precision transmission gear is running; the vibration data is obtained by an acceleration sensor to record the vibration condition of the high-precision transmission gear under a specified speed and a specified load; the surface image data is collected by a high-resolution industrial camera.

[0156] The front feature, the back feature, the sound feature and the vibration feature are standardized using the same standardization processing method to obtain the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature; the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature are subjected to feature fusion in the same manner as the multimodal feature fusion to generate a fusion feature;

[0157] The fusion features are input into the trained fault prediction model to obtain prediction results; the prediction results include good gears and faulty gears.

[0158] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A gear production and finishing processing system, characterized in that: include: Historical data acquisition module, used to obtain historical processing data of high-precision transmission gears; A real-time data acquisition module is used to obtain in real time the high-precision transmission gear surface image after the current processing stage is completed on the first production line to obtain a stage surface image; the processing stage includes a raw material preparation stage, a blank processing stage, a heat treatment stage and a finishing stage; A stage detection module, used to obtain stage real-time detection data according to the stage surface image; and obtain stage detection results according to the stage real-time detection data and corresponding standard detection data; A classification module is used to classify the high-precision transmission gears according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; remove the waste products from the first production line; and move the process-adjusted products and the reworked products to the second production line; A processing parameter adjustment module, for clustering the process-adjusted products and the reworked products according to the real-time detection data of the stages in each processing stage, respectively obtaining a plurality of processing batches and a plurality of rework batches; for each processing batch, obtaining a processing adjustment parameter according to the corresponding processing stage and the corresponding stage detection result; for each rework batch, determining a rework adjustment parameter according to the corresponding processing stage and the corresponding stage detection result; The second production line control module is used to adjust the second production line according to the processing adjustment parameters and the rework adjustment parameters respectively.

2. A gear production and finishing processing system according to claim 1, characterized in that: The historical processing data of the high-precision transmission gear is obtained as follows: The historical processing data includes the operating parameters of the high-precision transmission gear at each processing stage, and the detection data, detection results and post-processing data after each processing stage is completed; the detection results include qualified and unqualified; the post-processing data includes subsequent processing stages and subsequent processing parameters; Acquire the test data with qualified test results to obtain historical qualified data; analyze and extract key quality parameters based on the historical qualified data to generate the standard test data; the standard test data includes a set of key quality parameter ranges corresponding to each processing stage; the post-processing data of the historical qualified data is empty; The test data with unqualified test results is obtained to obtain historical unqualified data; the post-processing data of the historical unqualified data includes a subsequent stage and subsequent stage processing adjustment parameters.

3. A gear production and finishing processing system according to claim 2, characterized in that: According to the stage detection results, the high-precision transmission gears are classified as follows: Inputting the stage detection result into the trained corresponding stage classification model to obtain the stage classification result; The training process of the stage classification model includes the following steps: The historical qualified data and the historical unqualified data are used as training data; the training data is preprocessed, including missing value filling, outlier removal and data normalization, to obtain preprocessed training data; according to the detection results and the post-processing data, a stage label is assigned to each data sample in the preprocessed training data; the stage labels include qualified product labels, process adjustment product labels, rework product labels and waste product labels; the stage labels are used to train the stage classification model; feature engineering is performed according to the preprocessed training data to construct multiple surface size ratio features to obtain an extended feature set; principal component analysis is used to select key features from the extended feature set to obtain a selected feature set; the selected feature set is used as an independent variable, the stage label is used as a dependent variable, and the stage classification model is trained using a gradient boosting tree to obtain N stage classification models, where N is the number of the processing stages.

4. A gear production and finishing processing system according to claim 1, characterized in that: The surface image of the stage and the real-time detection data of the stage are specifically: The stage surface images include the front image of the raw material preparation stage, the back image of the raw material preparation stage, the front image of the blank processing stage, the back image of the blank processing stage, the front image of the heat treatment stage, the back image of the heat treatment stage, the front image of the finishing stage and the back image of the finishing stage; the stage real-time detection data includes the raw material preparation stage detection data, the blank processing stage detection data, the heat treatment stage detection data and the finishing stage detection data.

5. A gear production and finishing processing system according to claim 4, characterized in that: The real-time detection data of the stage obtained according to the surface image of the stage is specifically: For the front image of the raw material preparation stage and the back image of the raw material preparation stage, the following operations are performed: Preprocessing the front image of the raw material preparation stage and the back image of the raw material preparation stage to remove noise and uneven illumination, and obtain a standard raw material preparation image; extracting surface features and surface defect information from the standard raw material preparation image to obtain the raw material preparation stage detection data; the raw material preparation stage detection data includes surface roughness, the number of surface defects and surface color uniformity; For the front image of the blank processing stage and the back image of the blank processing stage, the following operations are performed: Preprocessing the front image of the blank processing stage and the back image of the blank processing stage to obtain a standard blank processing image; extracting the geometric contour and surface features of the blank from the standard blank processing image to obtain the blank processing stage detection data, wherein the blank processing stage detection data includes the outer diameter, length, width and thickness of the blank; For the front image at the heat treatment stage and the back image at the heat treatment stage, the following operations are performed: Preprocessing the front image of the heat treatment stage and the back image of the heat treatment stage to obtain a standard heat treatment image; using an image analysis algorithm to extract surface features from the standard heat treatment image to obtain the heat treatment stage detection data, wherein the heat treatment stage detection data includes a surface hardness value, a surface color change value, a surface crack length, a surface deformation amount, and a surface roughness; For the front image at the finishing stage and the back image at the finishing stage, the following operations are performed: The front image of the finishing stage and the back image of the finishing stage are preprocessed, including denoising, graying and edge enhancement, to obtain a standard finishing image; an edge detection algorithm is used to extract the outline of the gear from the standard finishing image, identify the key dimensions, and obtain an image processing result; the key dimensions include the tooth top circle, tooth root circle, tooth thickness, tooth pitch and tooth profile angle; according to the image processing result, the dimensional parameters of the high-precision transmission gear are calculated; the dimensional parameters include tooth height, tooth pitch, tooth thickness, gear inner diameter and outer diameter; according to the resolution and imaging ratio of the camera that collects the surface image of the stage, the dimensional parameters are calibrated, the size of the pixel unit is converted into the actual physical size, and the calibrated dimensional parameters are obtained; based on the extracted key dimensions and the calibrated dimensional parameters, the actual physical dimensional parameters of the high-precision transmission gear are calculated to obtain the finishing stage detection data.

6. A gear production and finishing processing system according to claim 1, characterized in that: For each of the processing batches, obtaining the processing adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the processing batch respectively to obtain the batch processing quantity; if the batch processing quantity is greater than the corresponding stage production quantity threshold, determining the processing adjustment parameters of the processing batch according to the historical processing data.

7. A gear production and finishing processing system according to claim 1, characterized in that: For each of the rework batches, determining the rework adjustment parameters according to the corresponding processing stage and the corresponding stage detection results also includes: counting the number of high-precision transmission gears in the rework batch respectively to obtain the rework quantity; if the rework quantity is greater than the corresponding stage rework production quantity threshold, determining the rework adjustment parameters of the rework batch according to the historical processing data.

8. A gear production and finishing processing system according to claim 1, characterized in that: The system further comprises a finishing multi-modal detection step, wherein the finishing multi-modal detection step comprises: Acquire the sound data, vibration data and surface image data of the high-precision transmission gear after the finishing stage, and acquire the historical sound data, historical vibration data, historical surface image data and annotation data of the high-precision transmission gear on the historical production line after the finishing stage; the surface image data and the historical surface image data both include the front image of the gear and the back image of the gear; the annotation data include the historical quality inspection results of the high-precision transmission gear, and the historical quality inspection results include faults and non-faults; The historical sound data, the historical vibration data and the historical surface image data are processed using a training set feature engineering process, including: extracting features of the front image of the gear and the back image of the gear in the historical surface image data respectively by a convolutional neural network to obtain historical front features and historical back features; extracting frequency domain features of the historical sound data by a time-frequency analysis method; extracting time domain features of the historical sound data and the historical vibration data by a long short-term memory network, and obtaining historical sound features and historical vibration features after integration; The extracted historical front features, historical back features, historical sound features, and historical vibration features are standardized to obtain historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features; the historical front standard features, historical back standard features, historical sound standard features, and historical vibration standard features are multimodally fused to generate historical fusion features; According to the historical fusion features and the labeled data, the LightGBM model is used for supervised learning to train a fault prediction model to identify whether a gear is faulty.

9. A gear production and finishing processing system according to claim 8, characterized in that: The system further comprises a finishing multi-modal detection step, wherein the finishing multi-modal detection step further comprises: Processing the sound data, the vibration data and the surface image data through the same steps as the training set feature engineering process, including: extracting front features and back features through the convolutional neural network, extracting frequency domain features of the sound data using the time-frequency analysis, extracting time domain features of the sound data and the vibration data using the long short-term memory network, and obtaining sound features and vibration features after integration; The front feature, the back feature, the sound feature and the vibration feature are standardized using the same standardization processing method to obtain the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature; the front standard feature, the back standard feature, the historical sound standard feature and the historical vibration standard feature are subjected to feature fusion in the same manner as the multimodal feature fusion to generate a fusion feature; The fusion features are input into the trained fault prediction model to obtain prediction results; the prediction results include good gears and faulty gears.

10. A gear production and finishing method, characterized in that: The following steps are involved: Obtain historical processing data of high-precision transmission gears; Real-time acquisition of high-precision transmission gear surface images after the current processing stage is completed on the first production line to obtain stage surface images; the processing stages include raw material preparation stage, blank processing stage, heat treatment stage and finishing stage; Obtaining stage real-time detection data according to the stage surface image; obtaining stage detection results according to the stage real-time detection data and corresponding standard detection data; Classify the high-precision transmission gears according to the stage detection results to obtain qualified products, process-adjusted products, reworked products and waste products; removing the waste products from the first production line; moving the process-adjusted products and the reworked products to the second production line; Clustering is performed based on the real-time detection data of the process-adjusted products and the reworked products at each of the processing stages to obtain a plurality of processing batches and a plurality of rework batches respectively; for each of the processing batches, a processing adjustment parameter is obtained based on the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted based on the processing adjustment parameter; for each of the rework batches, a rework adjustment parameter is determined based on the corresponding processing stage and the corresponding stage detection result; the second production line is adjusted based on the rework adjustment parameter.

Images