A bearing process parameter closed-loop regulation method and system based on real-time analysis

Abstract

The application relates to the technical field of intelligent manufacturing, and discloses a bearing process parameter closed-loop regulation and control method and system based on real-time analysis, which comprises the following steps: firstly, cross-time-window cross-correlation matching is carried out on a multi-dimensional time sequence working condition sequence and a quality deviation sequence, and a working condition-quality alignment sample set is constructed; a time-varying response surface and a transfer entropy matrix are built on the basis of the sample set, a dominant regulation and control parameter set is determined through joint analysis of time and space, the transfer entropy matrix is transformed and mapped, and a full-process causal chain topological graph is generated. Then, hierarchical dynamic adjustment is carried out on the dominant regulation and control parameter set, adaptive regulation and control instructions are generated, the quality deviation sequence is calculated across processes in combination with the causal chain topological graph, and directional compensation instructions are obtained. The two types of instructions are integrated into fusion regulation and control instructions and executed, full inspection and measurement are carried out on the process flow parts, the measured quality deviation is used to update the quality deviation sequence, and full-chain closed-loop regulation and control is formed; and the application can improve the stability of bearing machining quality.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a closed-loop control method and system for bearing process parameters based on real-time analysis. Background Technology

[0002] The control of bearing manufacturing process directly determines the precision grade and performance of bearing products, and is a core link in the production process of basic components for high-end equipment. Current traditional control methods lack precise time alignment and cross-parameter matching means in the data correlation stage, making it difficult to accurately establish the correspondence between process parameters and final quality. This leads to unclear positioning of quality-influencing factors, which in turn affects the effectiveness of process adjustments.

[0003] Existing technologies have significant shortcomings in process modeling and control execution. They fail to fully characterize the dynamic correlation characteristics of the process, only establishing static parameter and quality relationship models, making it difficult to adapt to time-varying production conditions. Furthermore, existing technologies cannot accurately identify dominant control parameters or clearly present the causal transmission relationships between different processes. The tracing and compensation of quality deviations are limited to a single process; control commands from different sources are prone to conflict and redundancy; the feedback mechanism is incomplete, failing to form a continuous iterative optimization system with a closed-loop chain. This results in significant fluctuations in product quality and insufficient production stability. Therefore, how to achieve dynamic and precise control of bearing process parameters has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a closed-loop control method and system for bearing process parameters based on real-time analysis to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, this invention provides a closed-loop control method for bearing process parameters based on real-time analysis, comprising:

[0006] S1. Perform cross-window cross-correlation matching on the multi-dimensional time-series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data.

[0007] S2. Based on the working condition-quality aligned sample set, construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set;

[0008] S3. Perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data.

[0009] S4. Based on the working condition-quality aligned sample set, perform hierarchical dynamic adjustment on the dominant control parameter set to obtain the adaptive control command of the working condition data;

[0010] S5. Based on the full-process causal chain topology diagram, perform cross-process calculations on the quality deviation sequence to obtain the directional compensation instruction for the quality deviation sequence;

[0011] S6. Perform causal integration on the directional compensation command and the adaptive control command to obtain the fusion control execution command of the working condition data. Perform full inspection feedback measurement on the process flow parts after executing the fusion control execution command, and transmit back the obtained quality measured deviation value of the process flow parts. Update the quality deviation sequence to form a full-chain closed-loop control.

[0012] In a preferred embodiment, the step of performing cross-window cross-correlation matching between the multidimensional time-series operating condition sequence and the quality deviation sequence of the operating condition data to obtain the operating condition-quality aligned sample set of the operating condition data includes:

[0013] Synchronous cross-time window sliding frame division is performed on the multidimensional time-series operating condition sequence and the quality deviation sequence to obtain the operating condition frame of the multidimensional time-series operating condition sequence and the quality deviation frame of the quality deviation sequence.

[0014] Based on the operating condition frame, the quality deviation frame is time-calibrated to obtain the time-delay aligned quality deviation frame of the quality deviation frame;

[0015] Cross-parameter pairing is performed between the operating condition frame and the time delay alignment quality deviation frame to obtain time delay alignment sample pairs of the time delay alignment quality deviation frame;

[0016] The time-delay aligned sample pairs are integrated and aggregated to obtain the condition-quality aligned sample set of the operating condition data.

[0017] In a preferred embodiment, constructing the time-varying response surface and transfer entropy matrix of the operating condition-quality aligned sample set based on the operating condition-quality aligned sample set includes:

[0018] The working condition-quality aligned sample set is divided into time windows to obtain the time-varying response slice set of the working condition-quality aligned sample set;

[0019] The response features of the time-varying response slice set are decoupled to obtain the condition-quality correlation tensor of the condition-quality aligned sample set;

[0020] The time-variable coefficient decomposition of the working condition quality correlation tensor is performed to obtain the time-varying quality offset of the working condition quality correlation tensor.

[0021] Using the time-varying mass offset as the Z-axis, time as the Y-axis, and process parameters as the X-axis, a time-varying response surface is constructed for the operating condition-mass aligned sample set.

[0022] The correlation information of the working condition-quality alignment sample set is extracted to obtain the inter-process information set of the working condition-quality alignment sample set;

[0023] The inter-process information set is transformed into a matrix to obtain the process relationship matrix of the inter-process information set;

[0024] Invalid associations are removed from the process relationship matrix to obtain the causal skeleton matrix of the process relationship matrix;

[0025] Redundancy delay elimination is performed on the causal skeleton matrix to obtain the transfer entropy matrix of the working condition-quality aligned sample set.

[0026] In a preferred embodiment, the step of performing spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the operating condition data, and transforming and mapping the transfer entropy matrix to obtain the full-process causal chain topology graph of the operating condition data includes:

[0027] Peak backtracking is performed on the time-varying response surface to obtain candidate control parameters for the time-varying response surface;

[0028] Based on the transfer entropy matrix, cross-time window causality verification is performed on the candidate control parameters to obtain the set of dominant control parameters of the operating condition data;

[0029] The transfer entropy matrix is ​​dual-normalized to obtain the transmission intensity matrix of the transfer entropy matrix;

[0030] Perform a row and column search on the conduction intensity matrix to obtain the process nodes of the conduction intensity matrix;

[0031] Weak correlation terms are removed from the conduction intensity matrix to obtain a sparse causal conduction skeleton of the conduction intensity matrix;

[0032] Using the sparsed causal transmission skeleton as the edge set and the process nodes as the node set, a full-process causal chain topology graph of the working condition data is generated.

[0033] In a preferred embodiment, the process node for performing row and column retrieval on the conduction intensity matrix to obtain the conduction intensity matrix includes:

[0034] The row and column identifiers of the conduction intensity matrix are extracted to obtain the potential process node set of the conduction intensity matrix;

[0035] The conduction intensity matrix is ​​summed by row and column intensities to obtain the row aggregation vector and column aggregation vector of the conduction intensity matrix. The elements in the row aggregation vector are the row aggregation intensity values, and the elements in the column aggregation vector are the column aggregation intensity values.

[0036] Based on the row aggregation vector and the column aggregation vector, the correlation of the potential process node set is measured to obtain the importance index of the potential process node set. The calculation formula of the importance index is as follows:

[0037] ;

[0038] In the formula, The index for the potential process node set is: The importance index of potential process nodes, index in the row aggregation vector The corresponding row polymerization strength value, index in the column aggregation vector The corresponding column of polymerization strength values, For the row aggregation vector, The maximum row aggregation strength value in the row aggregation vector. For the column aggregation vector, The maximum column aggregation strength value in the column aggregation vector. This is a preset zero-disturbance constant;

[0039] Filter out potential process nodes whose importance index is less than a preset threshold to obtain the process nodes of the conduction strength matrix.

[0040] In a preferred embodiment, the step of performing hierarchical dynamic adjustment of the dominant control parameter set based on the operating condition-quality aligned sample set to obtain adaptive control instructions for the operating condition data includes:

[0041] Based on the working condition-quality aligned sample set, the cooperative relationship of the dominant control parameter set is decoupled to obtain the single parameter control strength of the dominant control parameter.

[0042] Based on the single parameter control strength, the set of dominant control parameters is arranged in a sensitive hierarchy to obtain a hierarchical parameter control sequence chain;

[0043] By performing a historical operating condition retrieval on the hierarchical parameter control sequence chain, the layered fluctuation range of the hierarchical parameter control sequence chain is obtained.

[0044] Based on the aforementioned layered fluctuation range, the hierarchical parameter control sequence chain is dynamically optimized layer by layer to obtain the control step sequence of the hierarchical parameter control sequence chain.

[0045] Using the aforementioned control step size sequence as a correction benchmark, the hierarchical parameter control sequence chain is parametrically corrected to obtain the corrected hierarchical parameter control sequence chain.

[0046] The modified hierarchical parameter control sequence chain is encapsulated with control instructions to obtain the adaptive control instructions for the operating condition data.

[0047] In a preferred embodiment, the step of performing cross-process calculations on the quality deviation sequence based on the full-process causal chain topology to obtain the directional compensation instruction for the quality deviation sequence includes:

[0048] Reverse the causal direction of the full-process causal chain topology graph to obtain the reverse process dependency graph of the full-process causal chain topology graph;

[0049] Based on the reverse process dependency graph, the deviation source decomposition is performed on the quality deviation sequence to obtain the deviation contribution vector of the quality deviation sequence;

[0050] The compensation parameters of the deviation contribution vector are obtained by performing an inverse solution on the deviation contribution vector.

[0051] The independent compensation parameters are aligned in the compensation direction to obtain the directional compensation parameters of the independent compensation parameters.

[0052] The directional compensation parameters are encoded and converted to obtain the directional compensation instructions for the quality deviation sequence.

[0053] In a preferred embodiment, the step of causally integrating the directional compensation command and the adaptive control command to obtain the fused control execution command of the operating condition data includes:

[0054] Based on the full-process causal chain topology diagram, the conflict source analysis of the directional compensation instruction and the adaptive control instruction is performed to obtain the conflict relationship set of the directional compensation instruction and the adaptive control instruction;

[0055] The conflict relation set is subjected to paradigm standardization to obtain the conflict resolution paradigm of the conflict relation set;

[0056] Based on the conflict resolution paradigm, the directional compensation instruction and the adaptive control instruction are collaboratively modified to obtain a conflict resolution instruction pair between the directional compensation instruction and the adaptive control instruction;

[0057] The conflict resolution instructions are feature merged and fused to obtain the fused control execution instructions of the operating condition data.

[0058] In a preferred embodiment, the step of performing full inspection feedback measurement on the process flow parts after executing the fusion control execution command, transmitting back the obtained measured quality deviation value of the process flow parts, updating the quality deviation sequence, and forming a full-chain closed-loop control includes:

[0059] Multi-source sensor collaborative measurement is performed on the process flow parts after the fusion control execution command is executed to obtain the quality parameters of the process flow parts;

[0060] Based on the standard allowable error, the quality parameters are compared to obtain the measured quality deviation value of the process transfer parts;

[0061] The quality deviation sequence is updated based on the measured quality deviation value;

[0062] The updated quality deviation sequence replaces the current quality deviation sequence, triggering cross-time window cross-correlation matching for the next regulatory cycle, thus forming a closed-loop regulation across the entire chain.

[0063] To address the aforementioned problems, the present invention also provides a closed-loop control system for bearing process parameters based on real-time analysis, the system comprising:

[0064] The working condition-quality aligned sample construction module is used to perform cross-window cross-correlation matching between the multi-dimensional time series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data.

[0065] The correlation model construction module is used to construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set based on the working condition-quality aligned sample set.

[0066] The control parameter screening and causal topology generation module is used to perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and to transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data.

[0067] The hierarchical dynamic adjustment and adaptive control instruction generation module is used to perform hierarchical dynamic adjustment on the dominant control parameter set based on the working condition-quality aligned sample set, so as to obtain the adaptive control instruction of the working condition data.

[0068] The deviation tracing and directional compensation instruction generation module is used to perform cross-process calculations on the quality deviation sequence based on the full-process causal chain topology diagram to obtain the directional compensation instruction for the quality deviation sequence.

[0069] The instruction fusion, execution, and closed-loop feedback update module is used to perform causal integration on the directional compensation instruction and the adaptive control instruction to obtain the fusion control execution instruction of the working condition data, perform full inspection feedback measurement on the process flow parts after executing the fusion control execution instruction, transmit back the obtained measured quality deviation value of the process flow parts, update the quality deviation sequence, and form a full-chain closed-loop control.

[0070] Compared with the prior art, the present invention has the following beneficial effects:

[0071] 1. This invention constructs a condition-quality aligned sample set by performing cross-correlation matching between multidimensional time-series operating condition sequences and quality deviation sequences across time windows. Based on this, a time-varying response surface and a transfer entropy matrix are generated, comprehensively characterizing the time-varying correlation characteristics of the process and the information transmission laws between processes. Through spatiotemporal joint analysis, the dominant control parameter set is accurately identified, and the transfer entropy matrix is ​​transformed and mapped to generate a full-process causal chain topology diagram, clearly presenting the causal transmission paths and intensity differences between processes. This provides direction for precise control of process parameters, avoiding the blindness of empirical adjustments.

[0072] 2. This invention generates adaptive control commands by hierarchically and dynamically adjusting the set of dominant control parameters. Based on the causal chain topology diagram of the entire process, it generates directional compensation commands by cross-process calculation of the quality deviation sequence, achieving dynamic adaptation and adjustment of process parameters and systematic traceability compensation of quality deviations. By causal integration and fusion of the two types of control commands, conflicts and redundancies between commands are eliminated, ensuring consistency in control execution. Combined with real-time feedback from full inspection and measurement of measured quality deviation values ​​to update the quality deviation sequence, a closed-loop control system is formed, enabling continuous iterative optimization of control strategies and enhancing the stability of the production process and the consistency of product quality. Attached Figure Description

[0073] Figure 1 A flowchart illustrating a closed-loop control method for bearing process parameters based on real-time analysis, provided in an embodiment of the present invention;

[0074] Figure 2 A functional block diagram of a bearing process parameter closed-loop control system based on real-time analysis is provided in an embodiment of the present invention.

[0075] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0076] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0077] This application provides a closed-loop control method for bearing process parameters based on real-time analysis. The executing entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the closed-loop control method for bearing process parameters based on real-time analysis can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0078] Reference Figure 1 The diagram shown is a flowchart illustrating a closed-loop control method for bearing process parameters based on real-time analysis, according to an embodiment of the present invention. In this embodiment, the closed-loop control method for bearing process parameters based on real-time analysis includes:

[0079] S1. Perform cross-window cross-correlation matching on the multi-dimensional time-series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data.

[0080] In this embodiment of the invention, the step of performing cross-window cross-correlation matching between the multidimensional time-series operating condition sequence and the quality deviation sequence of the operating condition data to obtain the operating condition-quality aligned sample set of the operating condition data includes:

[0081] Synchronous cross-time window sliding frame division is performed on the multidimensional time-series operating condition sequence and the quality deviation sequence to obtain the operating condition frame of the multidimensional time-series operating condition sequence and the quality deviation frame of the quality deviation sequence.

[0082] Based on the operating condition frame, the quality deviation frame is time-calibrated to obtain the time-delay aligned quality deviation frame of the quality deviation frame;

[0083] Cross-parameter pairing is performed between the operating condition frame and the time delay alignment quality deviation frame to obtain time delay alignment sample pairs of the time delay alignment quality deviation frame;

[0084] The time-delay aligned sample pairs are integrated and aggregated to obtain the condition-quality aligned sample set of the operating condition data.

[0085] The multi-dimensional time-series processing sequence collected from the bearing grinding production line includes four continuously recorded physical quantities: spindle speed, feed rate, depth of cut, and vibration amplitude. The quality deviation sequence records the offset of the corresponding machined part from the standard values ​​in terms of roundness, surface roughness, and dimensional tolerance. When performing synchronous cross-time window sliding frame segmentation on the multi-dimensional time-series processing sequence and the quality deviation sequence, the two columns of data are simultaneously extracted with the same time window length and sliding step size. The time window length is set to cover the entire process from single-piece loading to grinding completion, and the sliding step size is set to the single-piece processing cycle time. After extraction, a series of processing frames and quality deviation frames are formed in chronological order. Each processing frame contains spindle speed curve segments, feed rate curve segments, depth of cut curve segments, and vibration amplitude curve segments within the same time period. Each quality deviation frame contains roundness offset, surface roughness offset, and dimensional tolerance offset within the same time period.

[0086] The process of obtaining time-delay aligned quality deviation frames by performing time-series calibration on quality deviation frames based on operating condition frames involves performing offset matching on the time axis for each quality deviation frame, its corresponding operating condition frame, and the two adjacent operating condition frames. By calculating the cross-correlation between each quality offset within the quality deviation frame and each process parameter curve segment within the operating condition frame point by point, the time offset corresponding to the maximum correlation is found. The time coordinates of the quality deviation frame are then shifted overall according to the time offset corresponding to the maximum correlation, resulting in a time-delay aligned quality deviation frame that strictly corresponds to the operating condition frame in terms of time sequence. Specifically, the cross-correlation is calculated by multiplying the sequence of individual quality offsets within the quality deviation frame and the sequence of individual process parameters within the operating condition frame point by point at each time offset position and summing the results. The summation constitutes a cross-correlation sequence. The time offset corresponding to the point with the largest absolute value in the cross-correlation sequence is the time delay value between that quality offset and that process parameter. The mode of the time delay values ​​between each quality offset and each process parameter is taken as the overall time offset of the quality deviation frame.

[0087] When performing cross-parameter pairing of working condition frames and time-delay aligned quality deviation frames to obtain time-delay aligned sample pairs, the unique identifier of the workpiece is used as the pairing basis. Working condition frames belonging to the same workpiece are combined with time-series calibrated time-delay aligned quality deviation frames. Each time-delay aligned sample pair contains a complete set of process parameter curves and a complete set of quality offset records for a workpiece. All time-delay aligned sample pairs are integrated and collected in the order of processing time to form a working condition-quality aligned sample set. Each sample in this working condition-quality aligned sample set completely preserves the precise time-dimensional alignment between the dynamic change history of process parameters during bearing grinding and the final quality offset result, which can be directly read and used in subsequent steps.

[0088] The beneficial effects are that by cross-window cross-correlation matching, the multi-dimensional time-series operating condition sequence and the quality deviation sequence are strictly aligned in the time dimension, eliminating the time delay interference between process parameter changes and quality response, making the correspondence between operating condition data and quality data accurate to the single-piece processing level, providing a high-quality aligned sample set for subsequent response surface construction and causal analysis, avoiding the correlation analysis bias caused by time misalignment, and improving the accuracy of the mapping relationship between process parameters and quality indicators.

[0089] S2. Based on the working condition-quality aligned sample set, construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set;

[0090] In this embodiment of the invention, constructing the time-varying response surface and transfer entropy matrix of the operating condition-quality aligned sample set based on the operating condition-quality aligned sample set includes:

[0091] The working condition-quality aligned sample set is divided into time windows to obtain the time-varying response slice set of the working condition-quality aligned sample set;

[0092] The response features of the time-varying response slice set are decoupled to obtain the condition-quality correlation tensor of the condition-quality aligned sample set;

[0093] The time-variable coefficient decomposition of the working condition quality correlation tensor is performed to obtain the time-varying quality offset of the working condition quality correlation tensor.

[0094] Using the time-varying mass offset as the Z-axis, time as the Y-axis, and process parameters as the X-axis, a time-varying response surface is constructed for the operating condition-mass aligned sample set.

[0095] The correlation information of the working condition-quality alignment sample set is extracted to obtain the inter-process information set of the working condition-quality alignment sample set;

[0096] The inter-process information set is transformed into a matrix to obtain the process relationship matrix of the inter-process information set;

[0097] Invalid associations are removed from the process relationship matrix to obtain the causal skeleton matrix of the process relationship matrix;

[0098] Redundancy delay elimination is performed on the causal skeleton matrix to obtain the transfer entropy matrix of the working condition-quality aligned sample set.

[0099] The process of dividing the working condition-quality aligned sample set into time-window segments to obtain the time-varying response slice set involves dividing all samples in the working condition-quality aligned sample set into multiple continuous and non-overlapping time segments along the time axis. Each time segment contains the process parameter curve data and quality offset data of all processed parts within that time period. The length of the time segment is set to an integer multiple of the single clamping and processing cycle of the workpiece. Each slice in the time-varying response slice set formed after the division corresponds to a complete record of the process parameter changes and quality response within a fixed time period. The process of decoupling the response features of the time-varying response slice set to obtain the working condition quality correlation tensor involves sequentially extracting the corresponding strength between the temporal variation pattern of each process parameter and the variation trend of each quality index offset within each time-varying response slice. The time series of spindle speed, feed rate, depth of cut, and vibration amplitude within each time-varying response slice are taken as one dimension, and the time series of roundness offset, surface roughness offset, and dimensional tolerance offset within each time-varying response slice are taken as another dimension. A multidimensional array is constructed by calculating the degree of covariance between each pair of dimensions at each time point. This multidimensional array is the working condition quality correlation tensor. Each element within the working condition quality correlation tensor records the response strength value between a specific process parameter and a specific quality offset within a specific slice.

[0100] The process of obtaining time-varying mass offset by performing time-variable coefficient decomposition on the working condition quality correlation tensor involves expanding the working condition quality correlation tensor along the time dimension, performing multidimensional decomposition on the process parameter vector and mass offset vector at each time point, keeping the type of mass offset constant during decomposition, and expressing the mass offset as a weighted combination of each process parameter. The weight coefficient of the weighted combination at each time point reflects the contribution of the corresponding process parameter to the mass offset at that moment. All weighted combination values ​​of the same mass offset at all time points are connected in chronological order to form the offset curve of the mass offset changing with time. This curve is the time-varying mass offset. When constructing a time-varying response surface with time-varying mass offset as the Z-axis, time as the Y-axis, and process parameters as the X-axis, the time process of the machining process is unfolded along the Y-axis in three-dimensional space, and four process parameters—spindle speed, feed rate, depth of cut, and vibration amplitude—are arranged along the X-axis. The magnitude of the time-varying mass offset is calibrated along the Z-axis. By connecting each time point in space with the time-varying mass offset data points corresponding to each process parameter coordinate, a continuous spatial surface is formed. This spatial surface is the time-varying response surface. The time-varying response surface intuitively presents the dynamic influence of each process parameter on the mass offset at different machining stages during the bearing grinding process.

[0101] The process of extracting correlation information from the working condition-quality aligned sample set to obtain the inter-process information set involves systematically analyzing the information transmission paths between adjacent processes in the bearing production line. This involves extracting the numerical relationship between the quality indicators output by the previous process and the process parameters input by the next process from the working condition-quality aligned sample set. The transmission direction and intensity between upstream quality changes and downstream working condition adjustments are recorded for each pair of adjacent processes. All records of transmission direction and intensity between adjacent processes are compiled to form the inter-process information set. When performing matrix transformation on the inter-process information set to obtain the process relationship matrix, all processing process numbers are used as the row and column coordinates of the matrix. The row coordinates correspond to the upstream process that sends the information, and the column coordinates correspond to the downstream process that receives the information. The element in the i-th row and j-th column of the matrix records the information transmission intensity value from the i-th process to the j-th process. The transmission intensity value between each pair of processes in the inter-process information set is filled into the corresponding matrix position. Unrecorded process pairs are filled with zero values. The resulting square matrix is ​​the process relationship matrix. The process of eliminating invalid associations from the process relationship matrix to obtain the causal skeleton matrix involves sequentially checking whether each non-zero element in the process relationship matrix corresponds to a process pair with a direct physical causal relationship in terms of process logic. Elements corresponding to process pairs without a direct physical causal relationship are set to zero, while the transmission strength values ​​between process pairs with direct physical interaction between upstream and downstream processes are retained. The resulting matrix is ​​the causal skeleton matrix. When eliminating redundancy and delays from the causal skeleton matrix to obtain the transfer entropy matrix, indirect associations in the causal skeleton matrix that are not directly and immediately transmitted from upstream processes but are indirectly transmitted through intermediate processes are identified. Matrix elements corresponding to these indirect associations are set to zero, retaining only the information strength values ​​of direct and immediate transmission from upstream processes to downstream processes. The resulting matrix is ​​the transfer entropy matrix. The transfer entropy matrix, in a matrix form, characterizes the direct direction and direct strength of causal relationship transmission between processes throughout the entire production line.

[0102] The beneficial effects are that by constructing a time-varying response surface, the dynamic influence of process parameters on quality offset is fully characterized along the time axis. At the same time, by constructing a transfer entropy matrix, the direct direction and intensity of causal relationship transmission between processes in the entire production line are characterized in matrix form. The instantaneous influence effect of process parameters and the information transmission structure between processes are expressed separately, providing an independent and non-interfering data foundation for conducting parameter sensitivity analysis and full-process causal traceability, and eliminating the information coupling confusion caused by the sharing of the same data expression between response analysis and causal analysis.

[0103] S3. Perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data.

[0104] In this embodiment of the invention, the step of performing spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the operating condition data, and transforming and mapping the transfer entropy matrix to obtain the full-process causal chain topology graph of the operating condition data includes:

[0105] Peak backtracking is performed on the time-varying response surface to obtain candidate control parameters for the time-varying response surface;

[0106] Based on the transfer entropy matrix, cross-time window causality verification is performed on the candidate control parameters to obtain the set of dominant control parameters of the operating condition data;

[0107] The transfer entropy matrix is ​​dual-normalized to obtain the transmission intensity matrix of the transfer entropy matrix;

[0108] Perform a row and column search on the conduction intensity matrix to obtain the process nodes of the conduction intensity matrix;

[0109] Weak correlation terms are removed from the conduction intensity matrix to obtain a sparse causal conduction skeleton of the conduction intensity matrix;

[0110] Using the sparsed causal transmission skeleton as the edge set and the process nodes as the node set, a full-process causal chain topology graph of the working condition data is generated.

[0111] The process node for performing row and column searches on the conduction intensity matrix to obtain the conduction intensity matrix includes:

[0112] The row and column identifiers of the conduction intensity matrix are extracted to obtain the potential process node set of the conduction intensity matrix;

[0113] The conduction intensity matrix is ​​summed by row and column intensities to obtain the row aggregation vector and column aggregation vector of the conduction intensity matrix. The elements in the row aggregation vector are the row aggregation intensity values, and the elements in the column aggregation vector are the column aggregation intensity values.

[0114] Based on the row aggregation vector and the column aggregation vector, the correlation of the potential process node set is measured to obtain the importance index of the potential process node set. The calculation formula of the importance index is as follows:

[0115] ;

[0116] In the formula, The index for the potential process node set is: The importance index of potential process nodes, index in the row aggregation vector The corresponding row polymerization strength value, index in the column aggregation vector The corresponding column of polymerization strength values, For the column aggregation vector, The maximum row aggregation strength value in the row aggregation vector. For the column aggregation vector, The maximum column aggregation strength value in the column aggregation vector. This is a preset zero-disturbance constant;

[0117] Filter out potential process nodes whose importance index is less than a preset threshold to obtain the process nodes of the conduction strength matrix.

[0118] The process of obtaining candidate control parameters by peak backtracking on the time-varying response surface involves scanning the response curve segment corresponding to each process parameter layer by layer along the time axis on the time-varying response surface. The location where the absolute value of the time-varying mass offset reaches a local maximum on the response curve segment is identified. The process parameter corresponding to the location of the local maximum is extracted as a candidate control parameter. Each local maximum represents a significant impact of the corresponding process parameter on the mass offset within a specific processing period. All significantly influential process parameters extracted along the time axis are summarized to form a set of candidate control parameters. During peak backtracking, the method for determining local maxima is to compare the amplitude of the time-varying mass offset at three adjacent time points along the time direction on the response curve of a single process parameter on the time-varying response surface. When the amplitude at the intermediate time point is simultaneously greater than the amplitudes at the previous and next time points, that intermediate time point is the location of the local maximum, and the process parameter corresponding to this location is the candidate control parameter.

[0119] The process of obtaining the dominant control parameter set by performing cross-time-window causal verification on candidate control parameters based on the transfer entropy matrix involves placing each candidate control parameter in the candidate control parameter set into the process causal relationship network described by the transfer entropy matrix to test its causal transmission capability. Specifically, the transfer entropy matrix is ​​searched for a transmission path starting from the process where the candidate control parameter is located and ending at the process where the quality offset is located. If the transfer entropy matrix contains such a transmission path and the transmission intensity value recorded on the transmission path is not zero, it indicates that the candidate control parameter has a causally verified direct or indirect influence on the quality offset, and it is retained and included in the dominant control parameter set. If the transfer entropy matrix does not contain such a transmission path or the transmission intensity value on the transmission path is zero, it indicates that the influence of the candidate control parameter on the quality offset lacks causal chain support, and it is excluded from the dominant control parameter set. After the cross-time-window causal verification is completed, the dominant control parameter set obtained only retains those process parameters that both show significant response on the time-varying response surface and are verified in the causal transmission relationship.

[0120] The process of obtaining the transmission intensity matrix by dual normalization of the transmission entropy matrix involves symmetrically scaling each element in the matrix using the sum of the rows and the sum of the columns of the transmission entropy matrix. First, the cumulative value of all elements in each row of the transmission entropy matrix is ​​calculated as the row normalization base. Then, the cumulative value of all elements in each column of the transmission entropy matrix is ​​calculated as the column normalization base. Next, for the element in the i-th row and j-th column of the transmission entropy matrix, the original value of the element is divided by the geometric mean of the row normalization base and the column normalization base as the denominator to obtain the normalized element value. The new matrix formed by scaling all elements using the row normalization base and the column normalization base is the transmission intensity matrix. Each element in the transmission intensity matrix records the relative intensity of causal transmission between processes and eliminates the bias caused by the difference in absolute magnitude between the row and column directions.

[0121] The process of extracting row and column identifiers from the conduction strength matrix to obtain the potential process node set involves directly taking all process identifiers corresponding to the row and column indices of the conduction strength matrix, including all processes that appear in the row or column coordinates into the potential process node set, with each process retaining only one identifier in the potential process node set. The process of summing the row and column strengths of the conduction strength matrix to obtain the row aggregation vector and column aggregation vector involves summing the element values ​​of all columns in each row of the conduction strength matrix, and filling the corresponding index position in the row aggregation vector with the summed result. Similarly, for each column of the conduction strength matrix, summing the element values ​​of all rows in that column, and filling the corresponding index position in the column aggregation vector with the summed result. Each row aggregation strength value in the row aggregation vector reflects the total causal output strength of the corresponding process as a causal initiator, and each column aggregation strength value in the column aggregation vector reflects the total causal input strength of the corresponding process as a causal receiver.

[0122] When measuring the relevance of a potential process node set based on row and column aggregation vectors to obtain an importance index, the index for the potential process node set is: Potential process nodes, extract the index from the row aggregation vector. Corresponding row polymerization strength value Extract the index from the column aggregation vector Corresponding column polymerization strength value First calculate the row polymerization strength value. Polymer strength value The product of the product and then the product divided by the row polymerization strength value. The square of the column polymerization strength value The sum of squares plus the zero-disturbance constant The square root of the result is then used to add 1 to the quotient, followed by the natural logarithm to obtain the first factor. Finally, the polymerization strength value is taken. Polymer strength value The sum divided by the maximum row aggregation strength value in the row aggregation vector The maximum column aggregation strength value in the column aggregation vector The ratio obtained by summing the first and second parts is used to obtain the second part factor by taking the hyperbolic tangent function of the ratio. Multiplying the first part factor by the second part factor yields the importance index of the potential process node. :

[0123] ;

[0124] In the formula, The index for the potential process node set is: The importance index of potential process nodes, index in the row aggregation vector The corresponding row polymerization strength value, index in the column aggregation vector The corresponding column of polymerization strength values, For the column aggregation vector, The maximum row aggregation strength value in the row aggregation vector. For the column aggregation vector, The maximum column aggregation strength value in the column aggregation vector. This is the preset zero-disturbance constant.

[0125] Zero-disturbance constant The value is taken as a positive number much smaller than the normal row and column polymerization strength values ​​to avoid the denominator being zero. When filtering out potential process nodes with importance indices lower than the preset threshold to obtain process nodes, the preset threshold is set as a fixed ratio of the interval length between the maximum and minimum importance indices. All potential process nodes with importance indices lower than the preset threshold are removed from the potential process node set, and the remaining potential process nodes are the process nodes that constitute the causal chain topology graph of the entire process.

[0126] When removing weak correlation terms from the conduction strength matrix to obtain a sparse causal conduction skeleton, the value of each element in the conduction strength matrix is ​​checked one by one. Elements with values ​​lower than the stripping threshold are set to zero. The stripping threshold is set to a fixed proportion of the average value of all non-zero elements in the conduction strength matrix. The zeroing operation only cuts off marginal causal relationships with too weak conduction strength, and retains those causal relationships with significant conduction strength and actual regulatory influence between processes. After removing weak correlation terms, the conduction strength matrix becomes a sparse causal conduction skeleton with most elements being zero and only significant conduction paths are retained. When generating a full-process causal chain topology graph using a sparse causal transmission skeleton as the edge set and process nodes as the node set, each process node is drawn as a vertex in the topology graph on a plane. All non-zero elements in the sparse causal transmission skeleton are traversed. When the element in the i-th row and j-th column of the sparse causal transmission skeleton is not zero, a directed edge is drawn from the i-th process node to the j-th process node. The direction of the directed edge is consistent with the direction from the row coordinate to the column coordinate in the sparse causal transmission skeleton. The thickness of the directed edge corresponds proportionally to the value of the element. The directed graph formed after all process nodes and all directed edges are drawn is the full-process causal chain topology graph. The full-process causal chain topology graph fully presents the path structure and intensity distribution of the causal relationship between each process in the entire bearing processing process.

[0127] The beneficial effects are as follows: by spatiotemporal joint analysis, candidate parameters that have a significant impact on quality shift are located from the time-varying response surface, and the causal verification of the transfer entropy matrix is ​​used to screen out the set of dominant control parameters that have both significant response and causal transmission support, so as to avoid including parameters that only have statistical correlation but no causal driving relationship in the control range. At the same time, by row and column retrieval and importance index calculation, the process nodes that constitute the causal topology are accurately extracted from the transmission intensity matrix and weak correlation items are eliminated. The generated full-process causal chain topology accurately reflects the path structure and intensity distribution of the real physical causal transmission between processes.

[0128] S4. Based on the working condition-quality aligned sample set, perform hierarchical dynamic adjustment on the dominant control parameter set to obtain adaptive control instructions for the working condition data.

[0129] In this embodiment of the invention, the step of performing hierarchical dynamic adjustment of the dominant control parameter set based on the operating condition-quality aligned sample set to obtain adaptive control instructions for the operating condition data includes:

[0130] Based on the working condition-quality aligned sample set, the cooperative relationship of the dominant control parameter set is decoupled to obtain the single parameter control strength of the dominant control parameter.

[0131] Based on the single parameter control strength, the set of dominant control parameters is arranged in a sensitive hierarchy to obtain a hierarchical parameter control sequence chain;

[0132] By performing a historical operating condition retrieval on the hierarchical parameter control sequence chain, the layered fluctuation range of the hierarchical parameter control sequence chain is obtained.

[0133] Based on the aforementioned layered fluctuation range, the hierarchical parameter control sequence chain is dynamically optimized layer by layer to obtain the control step sequence of the hierarchical parameter control sequence chain.

[0134] Using the aforementioned control step size sequence as a correction benchmark, the hierarchical parameter control sequence chain is parametrically corrected to obtain the corrected hierarchical parameter control sequence chain.

[0135] The modified hierarchical parameter control sequence chain is encapsulated with control instructions to obtain the adaptive control instructions for the operating condition data.

[0136] The process of decoupling the synergistic relationship of the dominant control parameter set based on the operating condition-mass aligned sample set to obtain the control strength of a single parameter involves performing independent perturbation analysis on each dominant control parameter in the dominant control parameter set within the operating condition-mass aligned sample set. Specifically, the independent perturbation analysis involves selecting a subset of samples from the operating condition-mass aligned sample set where all dominant control parameters except the one currently being analyzed remain essentially constant. Within this subset, the change in mass offset corresponding to the individual change of the dominant control parameter being analyzed is observed. The ratio of the change in mass offset to the change in the dominant control parameter is taken as the control strength of that single parameter. The control strength of a single parameter reflects the independent driving strength of a single dominant control parameter on the mass offset after eliminating the coupling effects of other parameters.

[0137] When arranging the set of dominant control parameters according to the sensitivity hierarchy to obtain the hierarchical parameter control sequence chain, all dominant control parameters in the set of dominant control parameters are sorted from largest to smallest according to their individual control strength. The dominant control parameter with the largest individual control strength is placed at the first level of the sequence chain, the dominant control parameter with the second largest individual control strength is placed at the second level of the sequence chain, and so on until the dominant control parameter with the smallest individual control strength is placed at the last level of the sequence chain. After sorting, a hierarchical parameter sequence with a strict order is formed. This hierarchical parameter sequence is the hierarchical parameter control sequence chain, which specifies the priority adjustment order of each dominant control parameter in subsequent control.

[0138] The process of obtaining stratified fluctuation ranges by associating historical operating conditions with the hierarchical parameter control sequence chain involves extracting the name of the dominant control parameter and its corresponding processing procedure identifier at each level of the hierarchical parameter control sequence chain. Using the processing procedure identifier and the name of the dominant control parameter as search criteria, a full search is performed on the operating condition-quality aligned sample set to retrieve all samples containing historical values ​​of the dominant control parameter. From the retrieved samples, the actual numerical range of the dominant control parameter in all historical processing processes is extracted. The minimum value is used as the lower boundary and the maximum value is used as the upper boundary to construct the stratified fluctuation range of the dominant control parameter. The stratified fluctuation range limits the allowable adjustment range of the corresponding dominant control parameter during the control process.

[0139] The process of obtaining the control step sequence by dynamically optimizing the hierarchical parameter control sequence chain based on the hierarchical fluctuation range is as follows: starting from the first level and proceeding downwards layer by layer according to the hierarchical order of the hierarchical parameter control sequence chain, multiple alternative adjustment amounts are tried within the corresponding hierarchical fluctuation range of the dominant control parameter to be controlled at the current level at a fixed step interval. For each alternative adjustment amount tried, the alternative adjustment amount is substituted into the sample closest to the current operating condition in the operating condition-quality alignment sample set to estimate the quality offset. The alternative adjustment amount that makes the quality offset estimate closest to zero, i.e., the ideal quality state, is selected as the control step of the dominant control parameter at that level. After the optimization of the current level is completed, the control step is fixed and the optimization continues to the next level until all levels in the hierarchical parameter control sequence chain have been optimized. The set of control step lengths arranged in hierarchical order after the optimization of all levels is completed is the control step sequence.

[0140] When modifying the hierarchical parameter control sequence chain using the control step size sequence as the correction benchmark to obtain the modified hierarchical parameter control sequence chain, the current setting value of the dominant control parameter of each level in the hierarchical parameter control sequence chain is added to the control step size of the corresponding level in the control step size sequence. The sum is used as the modified setting value of the dominant control parameter of that level. The modified setting value replaces the original current setting value in the hierarchical parameter control sequence chain. The new hierarchical parameter sequence formed after all levels are replaced is the modified hierarchical parameter control sequence chain. When the modified hierarchical parameter control sequence chain is encapsulated to obtain adaptive control instructions, the dominant control parameter name, corresponding processing step identifier, and modified set value of each level in the modified hierarchical parameter control sequence chain are arranged into a set of control instruction frames that can be directly read and executed by the production line control system according to the process execution order. Each instruction field in the control instruction frame contains three parts: target process number, target parameter code, and target set value. The complete control instruction frame formed by packaging all instruction fields is the adaptive control instruction. The adaptive control instruction is directly sent to the actuators of each process in the bearing production line to drive the machine tools to complete the parameter adjustment.

[0141] The beneficial effects are as follows: by decoupling the collaborative relationship, the coupling influence between the dominant control parameters is eliminated, and the independent control strength of each parameter is obtained. Based on the independent control strength, a hierarchical parameter control sequence chain is constructed to clarify the priority of each parameter adjustment. The control step size sequence is generated by dynamically optimizing layer by layer within the layered fluctuation range, so that the multi-parameter collaborative adjustment process is carried out in an orderly manner from the main to the secondary, preventing overshoot or oscillation caused by coupling effect when multi-parameter synchronous adjustment is performed. The adaptive control command is directly encapsulated into a command format that can be read by the production line actuator and issued to complete the parameter adjustment.

[0142] S5. Based on the full-process causal chain topology diagram, perform cross-process calculations on the quality deviation sequence to obtain the directional compensation instruction for the quality deviation sequence;

[0143] In this embodiment of the invention, the step of performing cross-process calculations on the quality deviation sequence based on the full-process causal chain topology to obtain the directional compensation instruction for the quality deviation sequence includes:

[0144] Reverse the causal direction of the full-process causal chain topology graph to obtain the reverse process dependency graph of the full-process causal chain topology graph;

[0145] Based on the reverse process dependency graph, the deviation source decomposition is performed on the quality deviation sequence to obtain the deviation contribution vector of the quality deviation sequence;

[0146] The compensation parameters of the deviation contribution vector are obtained by performing an inverse solution on the deviation contribution vector.

[0147] The independent compensation parameters are aligned in the compensation direction to obtain the directional compensation parameters of the independent compensation parameters;

[0148] The directional compensation parameters are encoded and converted to obtain the directional compensation instructions for the quality deviation sequence.

[0149] When reversing the causal direction of the entire process causal chain topology to obtain the reverse process dependency graph, each directed edge in the entire process causal chain topology is extracted, and the starting process node and ending process node of the directed edge are swapped. This changes the original directed edge pointing from process node i to process node j to a directed edge pointing from process node j to process node i. After all the directed edges are reversed, the node set of the entire process causal chain topology remains unchanged, but the direction of all edges is reversed. The directed graph formed after the reversal is the reverse process dependency graph. The reverse process dependency graph expresses the dependency path of tracing the impact of the deviation back along the process chain after the quality deviation occurs. Each directed edge in the graph points from the downstream process to the upstream process, indicating the direction of attribution of the quality defect.

[0150] The process of obtaining the deviation contribution vector by decomposing the quality deviation sequence based on the reverse process dependency graph is as follows: the final process output quality offset recorded in the quality deviation sequence is used as the starting point for tracing. The deviation is distributed upstream step by step along the reverse dependency path of the directed edge in the reverse process dependency graph. During the tracing distribution, the proportion of the transmission strength value between each process node in the reverse process dependency graph to the total transmission strength value of the entire tracing path is used as the distribution weight. The quality offset value of the final process output is distributed to each upstream process node through the tracing path according to the distribution weight. The part of the quality offset received by each upstream process node is the deviation contribution of that process. The vector formed by arranging the deviation contribution values ​​of all process nodes in the order of process number is the deviation contribution vector. Each element in the deviation contribution vector corresponds to the deviation amount that a process should be responsible for for the final quality deviation.

[0151] When performing inverse solving of the deviation contribution vector to obtain the independent compensation parameter, for each process in the deviation contribution vector, the correspondence between the process parameters and the quality offset recorded in the working condition-quality alignment sample set for that process is searched one by one. The corresponding records between the process parameter adjustment and the quality offset change of multiple sample points in the history of that process are extracted from the working condition-quality alignment sample set. In the corresponding records, the process parameter adjustment corresponding to the quality offset change that is equal to the absolute value of the current deviation contribution but opposite in direction is found. This process parameter adjustment is the independent compensation parameter required to offset the deviation contribution of that process to zero. The numerical direction of the independent compensation parameter is exactly opposite to the direction of the deviation contribution.

[0152] When aligning the compensation direction of independent compensation parameters to obtain directional compensation parameters, the direction of each process's independent compensation parameter is compared and verified with the deviation contribution of the same process in the deviation contribution vector. The direction in which the independent compensation parameter causes the quality offset to decrease is determined as the positive direction, and the direction in which the deviation contribution causes the quality offset to increase is determined as the negative direction. When the direction of the independent compensation parameter is opposite to the direction of the deviation contribution and the independent compensation parameter points in the positive direction, the independent compensation parameter is directly recorded as the directional compensation parameter. When the direction of the independent compensation parameter deviates from the direction of the deviation contribution after verification, the direction of the independent compensation parameter is flipped to meet the positive direction requirement of offsetting the deviation contribution before being recorded as the directional compensation parameter. The directional compensation parameters of all processes together constitute the process-level compensation quantity set used to compensate for the final quality deviation.

[0153] When encoding and converting directional compensation parameters to obtain directional compensation instructions, each directional compensation parameter is bound to its corresponding process identifier and process parameter identifier to form a compensation instruction entry. The compensation instruction entry contains four fields in sequence: target process number, name of process parameter to be adjusted, compensation adjustment amount value, and compensation execution direction identifier. The compensation execution direction identifier is positive to indicate that the compensation adjustment amount is increased based on the current setting value of the process parameter, and negative to indicate that the compensation adjustment amount is decreased based on the current setting value of the process parameter. All compensation instruction entries are sorted by topological distance from the final process in the reverse process dependency graph from farthest to near, and then concatenated to form a complete directional compensation instruction. After the directional compensation instruction is issued to the execution mechanism of each process on the production line, the corresponding compensation adjustment amount is superimposed on the current process parameter setting value to complete the directional compensation correction of quality deviation.

[0154] The beneficial effect is that by reversing the causal direction of the causal chain topology of the whole process to obtain the reverse process dependency graph, the final quality deviation is backtracked and distributed to each upstream process along the reverse dependency path to obtain the deviation contribution vector. Then, by combining the historical correspondence in the working condition-quality alignment sample set, the directional compensation parameters required to offset the deviation contribution of each process to zero are solved. This realizes a one-step reverse calculation from the final quality deviation result to the compensation amount of each process. The directional compensation command accurately specifies the direction and magnitude of the compensation adjustment amount of each process.

[0155] S6. Perform causal integration on the directional compensation command and the adaptive control command to obtain the fusion control execution command of the working condition data. Perform full inspection feedback measurement on the process flow parts after executing the fusion control execution command, and transmit back the obtained quality measured deviation value of the process flow parts. Update the quality deviation sequence to form a full-chain closed-loop control.

[0156] In this embodiment of the invention, the step of performing causal normalization on the directional compensation command and the adaptive control command to obtain the fused control execution command of the operating condition data includes:

[0157] Based on the full-process causal chain topology diagram, the conflict source analysis of the directional compensation instruction and the adaptive control instruction is performed to obtain the conflict relationship set of the directional compensation instruction and the adaptive control instruction;

[0158] The conflict relation set is subjected to paradigm standardization to obtain the conflict resolution paradigm of the conflict relation set;

[0159] Based on the conflict resolution paradigm, the directional compensation instruction and the adaptive control instruction are collaboratively modified to obtain a conflict resolution instruction pair between the directional compensation instruction and the adaptive control instruction;

[0160] The conflict resolution instructions are feature merged and fused to obtain the fused control execution instructions of the operating condition data.

[0161] The process involves performing full inspection and feedback measurement on the process flow parts after executing the fusion control command, transmitting the obtained measured quality deviation value of the process flow parts, updating the quality deviation sequence, and forming a full-chain closed-loop control, including:

[0162] Multi-source sensor collaborative measurement is performed on the process flow parts after the fusion control execution command is executed to obtain the quality parameters of the process flow parts;

[0163] Based on the standard allowable error, the quality parameters are compared to obtain the measured quality deviation value of the process transfer parts;

[0164] The quality deviation sequence is updated based on the measured quality deviation value;

[0165] The updated quality deviation sequence replaces the current quality deviation sequence, triggering cross-time window cross-correlation matching for the next regulatory cycle, thus forming a closed-loop regulation across the entire chain.

[0166] When performing conflict source analysis on directional compensation instructions and adaptive control instructions based on the full-process causal chain topology diagram to obtain the conflict relationship set, each process node in the full-process causal chain topology diagram is compared one by one. The direction and magnitude of the compensation adjustment amount specified by the directional compensation instruction for that process node, and the direction and magnitude of the correction adjustment amount specified by the adaptive control instruction for that process node are extracted respectively. When the adjustment directions specified by the two instructions on the same process node are completely consistent, the two instructions are determined to be non-conflicting and the process node is skipped. When the adjustment directions specified by the two instructions on the same process node are opposite, the process node is recorded as a conflicting process. The process identifier of the conflicting process, the adjustment direction and magnitude specified by the directional compensation instruction, the adjustment direction and magnitude specified by the adaptive control instruction, and the difference in the adjustment magnitude of the two instructions are recorded as a conflict entry. The set of entries formed by summarizing the conflict entries of all conflicting processes is the conflict relationship set.

[0167] When performing paradigm standardization on the conflict relation set to obtain the conflict resolution paradigm, each conflict entry in the conflict relation set is read one by one. Past processing instances with similar conflict states in the historical records of the conflict process are retrieved from the condition-quality alignment sample set. The retrieval criteria for past processing instances are that in the historical records, two instructions with opposite directions appear in the same process, and the difference in the amplitude of the two instructions falls within the neighborhood of the amplitude difference of the current conflict entry. The final adjustment amount actually used in history is extracted from the retrieved past processing instances. The final adjustment amount is then compared with the directional compensation instruction adjustment amount and the adaptive control instruction adjustment amount. The trade-off and compromise pattern in direction and magnitude is abstracted into a set of resolution rules. The resolution rules stipulate that when the adjustment quantities of two instructions in a conflicting process are opposite in direction, the direction of the one with the larger adjustment quantity magnitude is given priority as the fusion direction, and the weighted compromise value of the adjustment quantities magnitude of the two instructions is used as the fusion magnitude. The weight of the weighted compromise is taken as the proportion of the outward causal transmission strength value and the inward causal transmission strength value of the conflicting process node in the causal chain topology diagram of the whole process. Organizing this set of resolution rules into a standardized resolution format according to the process identifier of the conflicting process is called the conflict resolution paradigm. Each conflicting process corresponds to one conflict resolution paradigm.

[0168] When obtaining a conflict resolution instruction pair by coordinating the correction of directional compensation instructions and adaptive control instructions based on the conflict resolution paradigm, the compensation adjustment amount for each conflicting process in the directional compensation instruction is recalculated according to the conflict resolution paradigm corresponding to that conflicting process. The original compensation direction in the directional compensation instruction is replaced with the fusion direction specified in the conflict resolution paradigm, and the original compensation magnitude in the directional compensation instruction is replaced with the fusion magnitude specified in the conflict resolution paradigm. This yields the corrected instruction entry for that conflicting process in the directional compensation instruction. At the same time, the correction adjustment amount for the same conflicting process in the adaptive control instruction is also recalculated according to the same conflict resolution paradigm. The original correction direction in the adaptive control instruction is replaced with the fusion direction specified in the conflict resolution paradigm, and the original correction magnitude in the adaptive control instruction is replaced with the fusion magnitude specified in the conflict resolution paradigm. This yields the corrected instruction entry for that conflicting process in the adaptive control instruction. When the adjustment directions of the two instructions on the conflicting process are consistent, the original entries of the two instructions remain unchanged. The pair of corrected instruction entries formed after all conflicting processes have been processed constitutes the conflict resolution instruction pair.

[0169] When obtaining the fusion control execution command by feature merging and fusing conflict resolution command pairs, the directional compensation command entries and adaptive control command entries in the conflict resolution command pairs after collaborative correction are aligned one by one according to the process nodes. For each process node, when both compensation command entries and control command entries exist in the conflict resolution command pair, the adjustment amounts specified in the compensation command entry and the control command entry are added together, and the result is used as the fusion control amount for that process. The adjustment directions specified in the compensation command entry and the control command entry are superimposed in the same direction for confirmation, and the confirmed direction is used as the fusion control direction for that process. The four fields of process identifier, process parameter identifier, fusion control direction, and fusion control amount are combined into a fusion command entry. When the conflict resolution command pair contains only the compensation command entry or only the control command entry for that process, the single entry is directly promoted to a fusion command entry. The instruction set formed by arranging all the fusion command entries of all processes in the order of the production line process flow and packaging them together is the fusion control execution command.

[0170] When performing multi-source sensor collaborative measurement to obtain quality parameters for process flow parts after executing the fusion control command, multiple sets of sensors arranged at the exit of each process in the bearing production line simultaneously collect the key geometric and physical characteristics of the process flow parts. The roundness sensor outputs roundness measurement value by rotating the process flow part and recording the radial displacement runout. The surface roughness sensor outputs roughness measurement value by uniformly tracing the surface of the process flow part with a stylus and recording the vertical displacement change. The dimensional tolerance sensor outputs dimensional measurement value by non-contact laser ranging and recording the distance between the measured part of the process flow part and the reference position. The three types of sensors complete synchronous acquisition within the same measurement cycle and summarize their respective output values ​​into a set of quality parameters for the same process flow part.

[0171] When comparing quality parameters based on standard allowable errors to obtain the measured quality deviation value, the roundness measurement value of the process flow part is compared with the roundness qualified interval formed by the upper and lower limits of the roundness standard allowable error. The roundness deviation component is obtained by subtracting the center value of the roundness standard from the roundness measurement value. The surface roughness measurement value of the process flow part is compared with the roughness qualified interval formed by the upper and lower limits of the roughness standard allowable error. The roughness deviation component is obtained by subtracting the center value of the roughness standard from the roughness measurement value. The dimensional measurement value of the process flow part is compared with the dimensional qualified interval formed by the upper and lower limits of the dimensional standard allowable error. The dimensional deviation component is obtained by subtracting the center value of the dimensional standard from the dimensional measurement value. The roundness deviation component, roughness deviation component, and dimensional deviation component are arranged in the original order of the quality deviation sequence to form the measured quality deviation value.

[0172] When updating the quality deviation sequence based on the measured quality deviation value, the measured quality deviation value is appended to the end of the current quality deviation sequence as the quality offset record produced in the latest processing cycle. After appending, the length of the quality deviation sequence increases by one position, and the newly appended measured quality deviation value becomes the latest data point in the quality deviation sequence. The order and value of the original data in the quality deviation sequence remain unchanged. When the updated quality deviation sequence replaces the current quality deviation sequence to trigger the cross-time window cross-correlation matching of the next control cycle to form a full-chain closed-loop control, the updated quality deviation sequence is set as the current quality deviation sequence. The set current quality deviation sequence is then used to re-perform cross-time window cross-correlation matching with the newly acquired multi-dimensional time-series operating condition sequence to obtain a new operating condition-quality aligned sample set. Subsequently, the construction of the time-varying response surface and the transfer entropy matrix, the analysis of the dominant control parameter set and the causal chain topology of the whole process are carried out in sequence, the generation of adaptive control commands and directional compensation commands, and the merging and issuance of fusion control execution commands are carried out. The entire process from the input of the quality deviation sequence to the output of the fusion control command is executed cyclically. After each cycle is completed, the quality deviation sequence is updated based on the feedback-returned measured quality deviation value and the next cycle is started, thus forming a full-chain closed-loop control.

[0173] The beneficial effects are as follows: by analyzing the source of instruction conflicts, conflicting entries of directional compensation instructions and adaptive control instructions in the same process are identified. After the conflict resolution paradigm is used to collaboratively correct the conflicting instructions, they are merged and integrated to obtain a single fused control execution instruction. This eliminates the execution ambiguity and mutual interference that may occur when the two instructions are issued independently. At the same time, by using multi-source sensing collaborative measurement to collect the quality parameters of the process flow parts after the execution of the fused instruction, the quality deviation sequence is updated and fed back. This ensures that the quality deviation sequence always reflects the latest actual status of the production line. The cyclical execution mechanism of the whole-chain closed-loop control realizes continuous adaptive following between changes in working conditions and quality feedback.

[0174] like Figure 2 The diagram shown is a functional block diagram of a bearing process parameter closed-loop control system based on real-time analysis provided in an embodiment of the present invention.

[0175] The bearing process parameter closed-loop control system 100 based on real-time analysis described in this invention can be installed in an electronic device. Depending on the functions implemented, the bearing process parameter closed-loop control system 100 may include a working condition-quality alignment sample construction module 101, an association model construction module 102, a control parameter screening and causal topology generation module 103, a hierarchical dynamic adjustment and adaptive control instruction generation module 104, a deviation tracing and directional compensation instruction generation module 105, and an instruction fusion, execution, and closed-loop feedback update module 106. The modules described in this invention can also be called units, referring to a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0176] In this embodiment, the functions of each module / unit are as follows:

[0177] The working condition-quality aligned sample construction module 101 is used to perform cross-time window cross-correlation matching between the multidimensional time series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data.

[0178] The association model construction module 102 is used to construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set based on the working condition-quality aligned sample set.

[0179] The control parameter screening and causal topology generation module 103 is used to perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and to transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data.

[0180] The hierarchical dynamic adjustment and adaptive control instruction generation module 104 is used to perform hierarchical dynamic adjustment on the dominant control parameter set based on the working condition-quality aligned sample set, so as to obtain the adaptive control instruction of the working condition data.

[0181] The deviation tracing and directional compensation instruction generation module 105 is used to perform cross-process calculations on the quality deviation sequence based on the full-process causal chain topology diagram to obtain the directional compensation instruction for the quality deviation sequence.

[0182] The instruction fusion, execution, and closed-loop feedback update module 106 is used to perform causal correction on the directional compensation instruction and the adaptive control instruction to obtain the fusion control execution instruction of the working condition data, perform full inspection feedback measurement on the process flow parts after executing the fusion control execution instruction, transmit back the obtained quality measured deviation value of the process flow parts, update the quality deviation sequence, and form a full-chain closed-loop control.

[0183] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0184] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0185] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0186] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0187] The embodiments of this application can acquire and process relevant data based on an artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0188] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A closed-loop control method for bearing process parameters based on real-time analysis, characterized in that, The method includes: S1. Perform cross-window cross-correlation matching on the multi-dimensional time-series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data. S2. Based on the working condition-quality aligned sample set, construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set; S3. Perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data. S4. Based on the working condition-quality aligned sample set, perform hierarchical dynamic adjustment on the dominant control parameter set to obtain the adaptive control command of the working condition data; S5. Based on the full-process causal chain topology diagram, perform cross-process calculations on the quality deviation sequence to obtain the directional compensation instruction for the quality deviation sequence; S6. Perform causal integration on the directional compensation command and the adaptive control command to obtain the fusion control execution command of the working condition data. Perform full inspection feedback measurement on the process flow parts after executing the fusion control execution command, and transmit back the obtained quality measured deviation value of the process flow parts. Update the quality deviation sequence to form a full-chain closed-loop control.

2. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The step of performing cross-window cross-correlation matching between the multidimensional time-series operating condition sequence and the quality deviation sequence of the operating condition data to obtain the operating condition-quality aligned sample set of the operating condition data includes: Synchronous cross-time window sliding frame division is performed on the multidimensional time-series operating condition sequence and the quality deviation sequence to obtain the operating condition frame of the multidimensional time-series operating condition sequence and the quality deviation frame of the quality deviation sequence. Based on the operating condition frame, the quality deviation frame is time-calibrated to obtain the time-delay aligned quality deviation frame of the quality deviation frame; Cross-parameter pairing is performed between the operating condition frame and the time delay alignment quality deviation frame to obtain time delay alignment sample pairs of the time delay alignment quality deviation frame; The time-delay aligned sample pairs are integrated and aggregated to obtain the condition-quality aligned sample set of the operating condition data.

3. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The construction of the time-varying response surface and transfer entropy matrix of the operating condition-quality aligned sample set based on the operating condition-quality aligned sample set includes: The working condition-quality aligned sample set is divided into time windows to obtain the time-varying response slice set of the working condition-quality aligned sample set; The response features of the time-varying response slice set are decoupled to obtain the condition-quality correlation tensor of the condition-quality aligned sample set; The time-variable coefficient decomposition of the working condition quality correlation tensor is performed to obtain the time-varying quality offset of the working condition quality correlation tensor. Using the time-varying mass offset as the Z-axis, time as the Y-axis, and process parameters as the X-axis, a time-varying response surface is constructed for the operating condition-mass aligned sample set. The correlation information of the working condition-quality alignment sample set is extracted to obtain the inter-process information set of the working condition-quality alignment sample set; The inter-process information set is transformed into a matrix to obtain the process relationship matrix of the inter-process information set; Invalid associations are removed from the process relationship matrix to obtain the causal skeleton matrix of the process relationship matrix; Redundancy delay elimination is performed on the causal skeleton matrix to obtain the transfer entropy matrix of the working condition-quality aligned sample set.

4. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The process involves performing spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the operating condition data, and transforming and mapping the transfer entropy matrix to obtain the full-process causal chain topology graph of the operating condition data, including: Peak backtracking is performed on the time-varying response surface to obtain candidate control parameters for the time-varying response surface; Based on the transfer entropy matrix, cross-time window causality verification is performed on the candidate control parameters to obtain the set of dominant control parameters of the operating condition data; The transfer entropy matrix is ​​dual-normalized to obtain the transmission intensity matrix of the transfer entropy matrix; Perform a row and column search on the conduction intensity matrix to obtain the process nodes of the conduction intensity matrix; Weak correlation terms are removed from the conduction intensity matrix to obtain a sparse causal conduction skeleton of the conduction intensity matrix; Using the sparsed causal transmission skeleton as the edge set and the process nodes as the node set, a full-process causal chain topology graph of the working condition data is generated.

5. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 4, characterized in that, The process node for performing row and column searches on the conduction intensity matrix to obtain the conduction intensity matrix includes: The row and column identifiers of the conduction intensity matrix are extracted to obtain the potential process node set of the conduction intensity matrix; The conduction intensity matrix is ​​summed by row and column intensities to obtain the row aggregation vector and column aggregation vector of the conduction intensity matrix. The elements in the row aggregation vector are the row aggregation intensity values, and the elements in the column aggregation vector are the column aggregation intensity values. Based on the row aggregation vector and the column aggregation vector, the correlation of the potential process node set is measured to obtain the importance index of the potential process node set. The calculation formula of the importance index is as follows: ; In the formula, The index for the potential process node set is: The importance index of potential process nodes, index in the row aggregation vector The corresponding row polymerization strength value, index in the column aggregation vector The corresponding column of polymerization strength values, For the row aggregation vector, The maximum row aggregation strength value in the row aggregation vector. For the column aggregation vector, The maximum column aggregation strength value in the column aggregation vector. This is a preset zero-disturbance constant; Filter out potential process nodes whose importance index is less than a preset threshold to obtain the process nodes of the conduction strength matrix.

6. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The step of performing hierarchical dynamic adjustment of the dominant control parameter set based on the operating condition-quality aligned sample set to obtain adaptive control instructions for the operating condition data includes: Based on the working condition-quality aligned sample set, the cooperative relationship of the dominant control parameter set is decoupled to obtain the single parameter control strength of the dominant control parameter. Based on the single parameter control strength, the set of dominant control parameters is arranged in a sensitive hierarchy to obtain a hierarchical parameter control sequence chain; By performing a historical operating condition retrieval on the hierarchical parameter control sequence chain, the layered fluctuation range of the hierarchical parameter control sequence chain is obtained. Based on the aforementioned layered fluctuation range, the hierarchical parameter control sequence chain is dynamically optimized layer by layer to obtain the control step sequence of the hierarchical parameter control sequence chain. Using the aforementioned control step size sequence as a correction benchmark, the hierarchical parameter control sequence chain is parametrically corrected to obtain the corrected hierarchical parameter control sequence chain. The modified hierarchical parameter control sequence chain is encapsulated with control instructions to obtain the adaptive control instructions for the operating condition data.

7. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The step of performing cross-process extrapolation on the quality deviation sequence based on the full-process causal chain topology to obtain the directional compensation instruction for the quality deviation sequence includes: Reverse the causal direction of the full-process causal chain topology graph to obtain the reverse process dependency graph of the full-process causal chain topology graph; Based on the reverse process dependency graph, the deviation source decomposition is performed on the quality deviation sequence to obtain the deviation contribution vector of the quality deviation sequence; The compensation parameters of the deviation contribution vector are obtained by performing an inverse solution on the deviation contribution vector. The independent compensation parameters are aligned in the compensation direction to obtain the directional compensation parameters of the independent compensation parameters. The directional compensation parameters are encoded and converted to obtain the directional compensation instructions for the quality deviation sequence.

8. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The step of performing causal normalization on the directional compensation command and the adaptive control command to obtain the fused control execution command of the operating condition data includes: Based on the full-process causal chain topology diagram, the conflict source analysis of the directional compensation instruction and the adaptive control instruction is performed to obtain the conflict relationship set of the directional compensation instruction and the adaptive control instruction; The conflict relation set is subjected to paradigm standardization to obtain the conflict resolution paradigm of the conflict relation set; Based on the conflict resolution paradigm, the directional compensation instruction and the adaptive control instruction are collaboratively modified to obtain a conflict resolution instruction pair between the directional compensation instruction and the adaptive control instruction; The conflict resolution instructions are feature merged and fused to obtain the fused control execution instructions of the operating condition data.

9. The closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1, characterized in that, The process involves performing full inspection and feedback measurement on the process flow parts after executing the fusion control command, transmitting the obtained measured quality deviation value of the process flow parts, updating the quality deviation sequence, and forming a full-chain closed-loop control, including: Multi-source sensor collaborative measurement is performed on the process flow parts after the fusion control execution command is executed to obtain the quality parameters of the process flow parts; Based on the standard allowable error, the quality parameters are compared to obtain the measured quality deviation value of the process transfer parts; The quality deviation sequence is updated based on the measured quality deviation value; The updated quality deviation sequence replaces the current quality deviation sequence, triggering cross-time window cross-correlation matching for the next regulatory cycle, thus forming a closed-loop regulation across the entire chain.

10. A closed-loop control system for bearing process parameters based on real-time analysis, characterized in that, The system for implementing the closed-loop control method for bearing process parameters based on real-time analysis as described in claim 1 includes: The working condition-quality aligned sample construction module is used to perform cross-window cross-correlation matching between the multi-dimensional time series working condition sequence and the quality deviation sequence of the working condition data to obtain the working condition-quality aligned sample set of the working condition data. The correlation model construction module is used to construct the time-varying response surface and transfer entropy matrix of the working condition-quality aligned sample set based on the working condition-quality aligned sample set. The control parameter screening and causal topology generation module is used to perform spatiotemporal joint analysis on the time-varying response surface to obtain the dominant control parameter set of the working condition data, and to transform and map the transfer entropy matrix to obtain the full-process causal chain topology diagram of the working condition data. The hierarchical dynamic adjustment and adaptive control instruction generation module is used to perform hierarchical dynamic adjustment on the dominant control parameter set based on the working condition-quality aligned sample set, so as to obtain the adaptive control instruction of the working condition data. The deviation tracing and directional compensation instruction generation module is used to perform cross-process calculations on the quality deviation sequence based on the full-process causal chain topology diagram to obtain the directional compensation instruction for the quality deviation sequence. The instruction fusion, execution, and closed-loop feedback update module is used to perform causal integration on the directional compensation instruction and the adaptive control instruction to obtain the fusion control execution instruction of the working condition data, perform full inspection feedback measurement on the process flow parts after executing the fusion control execution instruction, transmit back the obtained measured quality deviation value of the process flow parts, update the quality deviation sequence, and form a full-chain closed-loop control.

Images