Dynamic compression and shear testing machine running state on-line monitoring and fault early warning system

Abstract

The application discloses a dynamic compression-shear testing machine operation state online monitoring and fault early warning system, belongs to the technical field of fault diagnosis, and aims to solve the problems of poor adaptability to multiple motion modes, lagging fault early warning and fuzzy fault positioning of the prior art. The system matches core cooperative link fault sensitive point monitoring parameters according to the current motion mode, generates a dynamic threshold to build a fault judgment threshold system, collects real-time data of a loading cycle division to generate a time series data set, judges the state of a candidate range based on a threshold judgment benchmark and builds a fault judgment benchmark library, predicts a trend through deviation analysis of the benchmark library and generates a compensation instruction, identifies potential faults, calculates link fault probability and completes diagnosis in combination with the threshold, actual value and predicted value. The application can improve monitoring adaptability and accuracy, realize early compensation and accurate fault early warning, and guarantee stable operation of equipment and test accuracy.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and more specifically to an online monitoring and fault early warning system for the operating status of a dynamic compression-shear testing machine. Background Technology

[0002] With the upgrading of industrial testing needs, mechanical performance testing machines need to adapt to various motion modes such as compression, shear, and mixed compression and shear, and higher requirements are placed on anomaly monitoring and fault early warning during operation. Currently, the industry mostly uses traditional technical solutions for monitoring the operating status and providing fault early warning for testing machines, which have revealed many problems that urgently need to be solved in practical applications. Existing monitoring solutions usually use fixed combinations of monitoring parameters and universal threshold standards, without taking into account the power transmission characteristics and the differences in the working of the core collaborative links under different motion modes for targeted configuration. This results in the monitoring parameters failing to accurately reflect the equipment's operating status under specific motion modes, and threshold determination is prone to misjudgment and omission, making it difficult to effectively identify early minor faults. At the same time, traditional solutions rely solely on the comparison between the real-time values ​​of monitoring parameters and preset thresholds to determine faults. They lack an initial benchmark that matches the current operating conditions of the equipment, and cannot capture the small deviations and evolution trends of monitoring parameters during loading cycles. They can only passively issue warnings after the parameters exceed the thresholds, which cannot intervene in the process of deviation accumulation in advance, easily leading to small deviations evolving into serious faults, and also causing a decrease in equipment operating accuracy, distortion of test data, and even downtime accidents. Therefore, in order to overcome these limitations, this invention proposes an online monitoring and fault early warning system for the operating status of a dynamic compression-shear testing machine. Summary of the Invention

[0003] To address the shortcomings of existing technologies, the present invention aims to provide an online monitoring and fault early warning system for the operating status of a dynamic compression-shear testing machine. This system solves the problems of adapting to different motion modes of the equipment, detecting equipment operating deviations in advance and making targeted adjustments, accurately locating faulty sub-links and identifying fault types, thereby ensuring stable equipment operation and testing accuracy.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] The online monitoring and fault early warning system for the operating status of the dynamic compression-shear testing machine includes a parameter configuration module, a benchmark construction module, a prediction and compensation module, and a fault identification module.

[0006] The parameter configuration module is used to match the monitoring parameters of the core collaborative link fault sensitive points according to the current motion mode and generate the dynamic threshold range of each monitoring parameter to build a fault judgment threshold system.

[0007] The benchmark construction module is used to collect real-time monitoring parameters and dynamic physical parameters of the dynamic compression-shear testing machine divided by loading cycle, generate a time-series monitoring dataset, and configure the benchmark candidate range. Based on the fault judgment threshold system, it determines whether the dynamic compression-shear testing machine in each loading cycle within the benchmark candidate range is in normal operation. If so, it triggers the fault judgment benchmark establishment operation and builds the fault judgment benchmark library.

[0008] The prediction and compensation module is used to perform deviation analysis on the time-series monitoring dataset based on the fault judgment benchmark library, predict the deviation trend, generate the predicted values ​​and confidence intervals of each monitoring parameter in the subsequent loading loop, and determine whether the compensation strategy generation process is triggered. If so, it generates compensation instructions according to the deviation trend type and deviation magnitude.

[0009] The fault identification module is used to identify potential fault monitoring parameters and construct a list of potential fault monitoring parameters based on the fault judgment threshold system, according to the actual values ​​of each monitoring parameter in the current loading cycle and the predicted values ​​of each monitoring parameter in the next loading cycle; and to identify faulty sub-links and perform fault type diagnosis by calculating the fault probability of the core collaborative sub-link.

[0010] Specifically, the dynamic threshold range includes a primary dynamic threshold range and a secondary dynamic threshold range;

[0011] The steps to construct a fault judgment threshold system include:

[0012] Before executing the test task, obtain the current motion mode of the dynamic compression-shear testing machine and select the core collaborative link that matches the current motion mode;

[0013] The core collaborative link is analyzed to extract the fault-sensitive points of the current core collaborative link; the fault-sensitive points are the key links of the core collaborative link that need to be monitored for faults.

[0014] Obtain monitoring parameters for fault-sensitive points in the core collaborative link, generate a list of monitoring parameters, and configure corresponding standard threshold ranges for each monitoring parameter.

[0015] Obtain historical normal data of the current core collaborative link monitoring parameters to calculate the distribution characteristic values ​​of each monitoring parameter, including the mean, standard deviation and confidence interval;

[0016] Based on the confidence interval in the distribution characteristic value, a reasonable fluctuation range of the monitoring parameters under the current actual operating state of the equipment is set; and according to the deviation between the reasonable fluctuation range and the standard threshold range, the standard threshold range boundary correction is performed to obtain the first-level dynamic threshold range of each monitoring parameter.

[0017] Match the link adaptation correction coefficient of the current core collaborative link, and further constrain the first-level dynamic threshold range to generate the second-level dynamic threshold range;

[0018] By associating and integrating the primary and secondary dynamic threshold ranges of each monitoring parameter, a fault judgment threshold system is constructed.

[0019] Specifically, the steps for generating a time-series monitoring dataset include:

[0020] Receive parameter configuration information matching the current motion mode, start the real-time data acquisition process, and obtain monitoring parameter data and dynamic physical parameters of the core collaborative link during the loading cycle of the dynamic compression and shear testing machine;

[0021] Based on the loading cycle control signal of the dynamic compression-shear testing machine, the start and end nodes of each loading cycle are identified, and the monitoring parameter data and dynamic physical parameters are segmented according to the loading cycle to form segmented data groups.

[0022] According to the loading cycle sequence, each segment of data is assigned a loading cycle sequence number identifier, which includes the loading cycle sequence code and the loading cycle start and end timestamps.

[0023] The monitoring parameter segment data under the same loading cycle number is bound to the dynamic physical parameter segment data, and validity is verified to generate a time-series monitoring dataset.

[0024] Specifically, the steps for constructing a fault diagnosis benchmark library include:

[0025] Configure the baseline candidate range, extract the loaded cyclic segmented data groups within the baseline candidate range from the time series monitoring dataset, and integrate them to form the baseline candidate dataset;

[0026] The fault judgment threshold system is invoked, and the first-level dynamic threshold range corresponding to each monitoring parameter is extracted as the basis for judging the running status. The candidate loop full normality detection process is executed to determine whether the loading loop is a normal loading loop.

[0027] If all loading loops within the baseline candidate range are determined to be normal loading loops, then the baseline candidate dataset is confirmed to be valid, triggering the fault judgment baseline establishment operation:

[0028] Based on the benchmark candidate dataset, feature values ​​of each core collaborative link are extracted. The feature values ​​include monitoring parameter feature values ​​and dynamic physical parameter feature values. At the same time, the inherent physical parameters of the core collaborative link are called as a supplement to the benchmark static attributes.

[0029] The feature dimensions are divided according to the core collaborative links. The feature values ​​of monitoring parameters and dynamic physical parameters under the same link are associated and bound with the inherent physical parameters to generate a subset of benchmark features corresponding to each core collaborative link.

[0030] The baseline feature subsets of all core collaborative links are integrated to form the fault judgment baseline library corresponding to this motion mode.

[0031] Specifically, the steps for predicting the deviation trend and generating predicted values ​​and confidence intervals for each monitoring parameter in the subsequent loading cycle include:

[0032] Continuously receive segmented data groups of subsequent loading loops from the time-series monitoring dataset; call the first-level dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system to determine whether the current loading loop is a normal loading loop; if so, trigger the deviation analysis operation.

[0033] Extract the monitoring parameters and dynamic physical parameters that are determined to be normal loading cycles; at the same time, backtrack the historical normal loading cycle segment data groups of the preset monitoring period before the current loading cycle, and extract the corresponding historical monitoring parameters and historical dynamic physical parameters;

[0034] Perform smoothing filtering preprocessing and data alignment operations on the segmented data groups of the current loading loop and historical normal loading loops, and integrate them to form a dataset for predictive analysis;

[0035] Call the fault judgment benchmark library and match the corresponding benchmark feature subset according to the core collaborative link dimension; based on the dataset for predictive analysis, perform deviation data calculation, the deviation data includes instantaneous deviation, cumulative deviation within the loop and deviation change rate, forming a deviation analysis matrix;

[0036] The deviation data of the current loading loop and the historical normal loading loop are extracted from the deviation analysis matrix and arranged in chronological order to form a historical deviation time series dataset.

[0037] The pre-built deviation trend prediction model is invoked, and the processed historical deviation time series dataset is substituted into the deviation trend prediction model. The predicted deviation values ​​and confidence intervals of the monitored parameters in the subsequent loading loop are output.

[0038] Specifically, the deviation trend prediction model consists of a data input layer, a bidirectional LSTM feature extraction layer, an attention weight allocation layer, a fully connected prediction layer, and an output layer.

[0039] The data input layer is used to receive historical deviation time-series datasets;

[0040] A bidirectional LSTM feature extraction layer is used to mine the bidirectional temporal correlation features of the deviation data in the historical deviation time series dataset, and obtain the deviation feature matrix;

[0041] The attention weight allocation layer is used to perform differential weight allocation on the bias feature matrix of the bidirectional LSTM output to obtain a weighted feature vector;

[0042] The fully connected prediction layer is used to perform non-linear transformation and dimension mapping on the weighted feature vector output by the attention weight allocation layer, and to map the weighted feature vector output by the attention layer to the bias prediction value of the subsequent loading loop.

[0043] The output layer integrates the prediction results of the fully connected layers, outputting the bias prediction value and its confidence interval.

[0044] Specifically, the steps for generating compensation instructions based on the type and magnitude of the deviation trend include:

[0045] Based on the values ​​of each monitoring parameter in the current loading cycle and the predicted deviation and confidence interval of the monitoring parameters in the next loading cycle, the predicted values ​​and confidence intervals of each monitoring parameter in the next loading cycle are generated.

[0046] The dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system is invoked to determine the deviation status of each monitoring parameter; the deviation status includes no compensation requirement, ordinary compensable deviation, critical compensable deviation and uncompensable fault deviation.

[0047] For monitoring parameters that are determined to be ordinary compensable deviations or critical compensable deviations, multi-dimensional attribute information is extracted based on the deviation analysis matrix, including link affiliation attribute, deviation magnitude attribute and trend evolution attribute, and integrated into standardized deviation attribute labels.

[0048] The system calls a preset compensation strategy library, which contains three-dimensional mapping rules for link affiliation attributes, deviation magnitude attributes, and trend evolution attributes, and is used to match compensation strategies based on multi-dimensional attribute information.

[0049] The matched compensation strategy is transformed into an executable standardized compensation instruction and sent to the corresponding execution agency to carry out the compensation operation;

[0050] After the next loading cycle is completed, the compensated monitoring parameters are extracted; if the compensated monitoring parameters are within the range of the secondary dynamic threshold, the compensation is deemed effective; if the deviation of the compensated monitoring parameters is outside the range of the secondary dynamic threshold, a targeted compensation strategy is regenerated and executed.

[0051] The number of times the targeted compensation strategy is regenerated and executed is counted. If the number exceeds the preset control threshold, the compensation is deemed to have failed, and an abnormal compensation warning is issued.

[0052] Specifically, the steps for identifying potential fault monitoring parameters and constructing a list of potential fault monitoring parameters include:

[0053] Receive the actual value dataset of each monitoring parameter in the current loading loop, as well as the predicted value and confidence interval of each monitoring parameter in the next loading loop; and call the fault judgment threshold system corresponding to the current motion mode, and associate the link affiliation identifier and fault sensitive point location information of each parameter;

[0054] The data is associated and bound according to the structure of monitoring parameter ID, link affiliation identifier, fault sensitive point location, value type, and dynamic threshold range to generate a judgment dataset.

[0055] Based on the link affiliation identifier of the judgment dataset, the monitoring parameters are divided into multiple core collaborative sub-link parameter groups according to the core collaborative link. The monitoring parameter attributes of the corresponding fault sensitive points are further associated within each core collaborative sub-link parameter group.

[0056] For the monitoring parameters of each core collaborative sub-link parameter group after splitting, a coupled determination of the actual value exceeding the limit status and the predicted value exceeding the limit trend is performed for each monitoring parameter, including:

[0057] If the actual value of a monitoring parameter in the current loading loop exceeds the corresponding first-level dynamic threshold range, it is determined to be a potential fault monitoring parameter, and the over-limit type is marked as actual value over-limit;

[0058] If the actual value of a monitoring parameter in the current loading cycle does not exceed the first-level dynamic threshold range, but the predicted value in the subsequent loading cycle exceeds the first-level dynamic threshold range, or the confidence interval intersects with the first-level dynamic threshold range, then a trend tracing judgment is initiated. Based on the temporal change slope of the historical actual value data of the monitoring parameter in the historical loading cycles, it is determined whether the monitoring parameter is a potential fault monitoring parameter. If so, the excess type is marked as predicted value excess.

[0059] A list of potential fault monitoring parameters is constructed based on the monitoring parameter ID, link affiliation identifier, fault sensitive point location, over-limit type, and over-limit magnitude of the potential fault monitoring parameters.

[0060] Specifically, the steps for identifying faulty sub-links and performing fault type diagnosis include:

[0061] Based on the list of potential fault monitoring parameters, physical features are extracted from each potential fault monitoring parameter to form a link fault feature evidence set; the physical features include core anomaly features and link-related features.

[0062] The extracted physical features are structured and integrated according to the structure of core collaborative sub-links, fault-sensitive points, physical feature types, and physical feature quantification values ​​to generate feature evidence subsets corresponding to each core collaborative sub-link.

[0063] By combining the power transmission logic of the core collaborative link with the fault causal relationship data stored in the fault knowledge base, a link-based fault causal graph is constructed.

[0064] The nodes of the link-based fault cause-effect graph include core collaborative sub-link nodes, fault-sensitive point nodes, and core component nodes; the directed edges between nodes represent the fault propagation relationship, and the weight of the edge represents the propagation probability.

[0065] The link failure feature evidence set is input into the constructed link-based failure causal graph, and the final failure probability of each core collaborative sub-link is calculated using a causal inference algorithm.

[0066] Set a failure probability threshold, and identify core collaborative sub-links whose final failure probability is higher than the failure probability threshold as failure sub-links, and form a list of failure sub-links;

[0067] If the list of faulty sub-links is not unique, then the operation of distinguishing between primary and secondary faulty links is performed based on the fault propagation origin, the timing information of the potential fault monitoring parameters corresponding to the faulty sub-link, or the fault probability.

[0068] The structured feature evidence subset of each faulty sub-link is matched with the fault type template in the fault knowledge base. Candidate fault types in the fault knowledge base are filtered according to the faulty sub-link affiliation. The matching degree between the feature evidence subset and each candidate fault type template is calculated to identify the final fault type.

[0069] Specifically, the steps for calculating the final failure probability of each core collaborative sub-link include:

[0070] Using the physical characteristics of each potential fault monitoring parameter as input, the triggering conditions of the corresponding fault-sensitive node in the link-based fault cause-effect graph are matched, and the node triggering probability is calculated.

[0071] Based on the preset propagation probability of the directed edge between the fault-sensitive node and the core collaborative sub-link node, the comprehensive fault probability of a single path of the core collaborative sub-link corresponding to a single fault-sensitive node is recursively calculated along the fault propagation direction, and used as the contribution value of the potential fault monitoring parameter to the propagation probability of the corresponding core collaborative sub-link.

[0072] The propagation probability contribution value refers to the quantitative contribution of the trigger probability of a fault-sensitive node corresponding to a certain potential fault monitoring parameter within the same core collaborative sub-link, after being weighted by the fault propagation probability between nodes, to the final failure probability of that core collaborative sub-link.

[0073] By combining the node trigger probability and propagation probability contribution values ​​corresponding to all potential fault monitoring parameters within the same core collaborative sub-link, the final fault probability of each core collaborative sub-link is obtained.

[0074] The beneficial effects of this invention are:

[0075] This application effectively solves the problems of poor parameter adaptability to multiple motion modes and misjudgment / missed judgment caused by the generalization of thresholds in traditional monitoring schemes by constructing dynamic threshold configuration based on motion modes and a working condition-specific benchmark library. This improves the accuracy of monitoring parameters in representing the equipment's operating status. By leveraging benchmark-driven deviation trend prediction and hierarchical compensation strategies, it achieves early perception and precise control of equipment operating deviations, solving the pain points of delayed fault warnings and lack of effective deviation pre-control mechanisms in traditional schemes. This prevents small deviations from accumulating into serious faults, ensuring the stability of equipment operation and the accuracy of test data. Through fault identification rules that couple actual and predicted values ​​and a link-based causal reasoning diagnostic method, it accurately locates faulty sub-links and distinguishes between primary and secondary links of compound faults. This solves the problems of vague fault location and low troubleshooting efficiency in traditional diagnostic schemes, significantly shortens fault handling time, reduces equipment operation and maintenance costs, and comprehensively improves the accuracy of testing machine operating status monitoring, the foresight of fault warnings, and the efficiency of fault diagnosis. Attached Figure Description

[0076] Figure 1 This is a schematic diagram of the online monitoring and fault early warning system for the dynamic compression-shear testing machine of the present invention;

[0077] Figure 2 A flowchart illustrating the construction of a fault diagnosis benchmark library for this invention;

[0078] Figure 3 This invention generates a flowchart of the predicted deviation values ​​and confidence intervals of the monitoring parameters to predict the deviation trend.

[0079] Figure 4 This is a flowchart illustrating how the present invention identifies faulty sub-links and performs fault type diagnosis. Detailed Implementation

[0080] Please see Figure 1 This embodiment introduces an online monitoring and fault early warning system for the operating status of a dynamic compression-shear testing machine, including a parameter configuration module, a benchmark construction module, a prediction compensation module, a fault identification module, and an early warning output module.

[0081] The parameter configuration module is used to determine the core collaborative link based on the motion mode of the dynamic compression-shear testing machine, and to complete the precise adaptation configuration of monitoring parameters and dynamic thresholds. The implementation process is as follows: Before executing the test task, based on the association mapping library between the motion mode and the core collaborative link, a core collaborative link matching the power transmission characteristics and working principle of the current motion mode is selected. Monitoring parameters of fault-sensitive points of the core collaborative link are obtained. Abandoning the fixed threshold configuration method, a dynamic threshold generation strategy combining the physical laws of the link and historical normal data statistics is adopted. A preset link adaptation correction coefficient is introduced to optimize the dynamic threshold range, generating a fault judgment threshold system. The dynamic threshold range includes a primary dynamic threshold range and a secondary dynamic threshold range. Simultaneously, parameter configuration information matching the current motion mode is output, including monitoring parameter type, acquisition frequency, dynamic threshold range, and link adaptation correction coefficient, providing a basis for subsequent monitoring, prediction, and fault judgment.

[0082] In this embodiment, the core of the parameter configuration module lies in solving the problems of poor compatibility between monitoring parameters and multiple motion modes of dynamic compression-shear testing machines, and the misjudgment and missed judgment caused by the generalization of thresholds in the prior art. By accurately matching the core collaborative link and monitoring parameters through the association mapping library, dynamic thresholds are generated by combining the physical laws of the link and historical data, and adaptation correction coefficients are introduced for optimization. Finally, a targeted fault judgment threshold system is formed, which not only ensures that the monitoring parameters can directly reflect the link operation status, but also provides a precise and unified parameter configuration basis for subsequent data acquisition, deviation prediction and fault judgment, thereby improving the adaptability and initial judgment accuracy of the entire monitoring and early warning system. The motion mode refers to the action mode in which the dynamic compression-shear testing machine applies different mechanical loads, including compression mode, shear mode, and mixed compression-shear mode. The core collaborative link refers to the hydraulic sub-link, mechanical sub-link, and control sub-link required for the dynamic compression-shear testing machine to achieve a specific motion mode. The association mapping library is a database that stores the correspondence between motion modes and core collaborative links. It is configured through a combination of offline sorting and online calibration based on the power transmission characteristics, link working principles, and historical application data of each motion mode of the dynamic compression-shear testing machine. For example, in compression mode, the core collaborative link is a combination of hydraulic sub-link, electrical control sub-link, and mechanical sub-link. The hydraulic sub-link includes the main hydraulic pump, relief valve, booster cylinder, and high-pressure oil pipe; the electrical control sub-link includes the PLC pressure controller, pressure sensor, and displacement sensor; and the mechanical sub-link includes the guide column, pressure platform, and limit device. The corresponding monitoring parameters cover the main oil circuit pressure, booster circuit pressure, flow rate, and oil temperature of the hydraulic system; the control voltage, signal response delay, and actuator operation frequency of the electrical control system; and the guide column clearance, pressure platform displacement, and vibration amplitude of the mechanical system. Fault-sensitive points are the key links in the core collaborative link that require fault monitoring. These are the critical links in the core collaborative link that are prone to failure and directly affect operational accuracy, ensuring that the monitoring parameters directly reflect the link's operational status. Monitoring parameters refer to physical and electrical parameters that characterize the operational status of each fault-sensitive point in the core collaborative link of the dynamic compression-shear testing machine, directly reflecting the link's performance and abnormal trends. These include hydraulic system parameters, mechanical mechanism parameters, and electrical control system parameters. Hydraulic system parameters include pressure, flow rate, and oil temperature; mechanical mechanism parameters include displacement, angle, vibration amplitude, and operating clearance; and electrical control system parameters include control voltage, current, signal response delay, and actuator operating frequency. The link adaptation correction coefficient is a preset threshold correction parameter based on the characteristics of different core collaborative links, used to improve the adaptability of the threshold to the link status.

[0083] Preferably, the specific steps for generating the fault judgment threshold system include:

[0084] Before executing the test task, the motion mode of the current dynamic compression and shear testing machine is obtained, the preset motion mode and core collaborative link association mapping library is called, and the collaborative link matching process is executed based on the power transmission characteristics and link working principle of the current motion mode to select the core collaborative link that is precisely matched with the current motion mode.

[0085] The core collaborative link is analyzed to extract the power transmission path parameters, key component model parameters, and correlation matrices between these parameters, generating a link analysis dataset. Based on this dataset, the fault-sensitive point identification rule base is invoked to execute the fault-sensitive point localization process and extract the fault-sensitive points of the current core collaborative link. The fault-sensitive point identification rule base is a database storing rules for locating fault-sensitive points in the core collaborative link. These rules are generated based on the structural characteristics, fault evolution mechanisms, and historical fault cases of the core collaborative link.

[0086] The system acquires monitoring parameters for fault-sensitive points in the core collaborative link, generates a monitoring parameter list, and calls a preset standard threshold mapping library. Based on the current core collaborative link type and monitoring parameter attributes, it executes a standard threshold configuration process to configure the corresponding standard threshold range for each monitoring parameter. The standard threshold mapping library is a database that stores the correspondence between core collaborative link types, monitoring parameter attributes, and standard threshold ranges. For example, the standard threshold range for the hydraulic circuit pressure monitoring parameter of the main oil circuit link in the hydraulic sub-link is 10MPa to 30MPa. The configuration method is as follows: initial configuration is completed after offline working condition verification, based on the design specifications of each core collaborative link of the dynamic compression-shear testing machine, the physical characteristics of the monitoring parameters, and industry-standard practices. Subsequent configuration is dynamically optimized and updated through an online calibration process.

[0087] The process involves acquiring historical normal data for the current core collaborative link monitoring parameters. Historical normal data refers to monitoring parameter data consistent with the current motion mode and core collaborative link configuration when the dynamic compression-shear testing machine is in normal operating condition. A data cleaning process is then performed on the historical normal data to remove abnormal interference data, resulting in a valid dataset. Based on this valid dataset, the distribution characteristic values ​​of each monitoring parameter are calculated, including the data mean, standard deviation, and confidence interval. A standard threshold adaptation adjustment process is then executed based on the distribution characteristic values. The specific adjustment logic steps are as follows: First, based on the confidence interval in the distribution characteristic values, a reasonable fluctuation range for the monitoring parameters under the current actual operating condition of the equipment is determined. Second, the deviation between this reasonable fluctuation range and the standard threshold range is calculated. Finally, based on the deviation, threshold boundary correction is performed, synchronously shifting the upper and lower boundaries of the standard threshold range according to the deviation, correcting it to a first-level dynamic threshold range that adapts to the current actual operating condition of the equipment. This addresses the problem of monitoring parameter benchmark deviation after prolonged use of the dynamic compression-shear testing machine, leading to a mismatch between the standard threshold range and the actual normal operating condition.

[0088] The system calls a pre-defined link adaptation correction coefficient library and matches the corresponding link adaptation correction coefficients based on the characteristic differences of the current core collaborative links. It then further constrains the first-level dynamic threshold range by combining the inherent characteristic differences of the core collaborative links, improving the adaptability of the threshold range to the link operating status and reducing false fault judgments caused by link characteristic differences. Finally, it executes a threshold correction process, substituting the link adaptation correction coefficients into the first-level dynamic threshold range and adjusting the upper and lower boundaries of the first-level dynamic threshold range through weighted calculation using the link adaptation correction coefficients to generate a second-level dynamic threshold range. The specific steps include:

[0089] The link adaptation correction coefficient is obtained from the library according to the core collaborative link type and motion mode. The value range of the link adaptation correction coefficient is 0.75-0.98. Its setting is determined based on the inherent physical characteristics of the core collaborative link, historical data accumulated over a long period of operation and maintenance, and the results of multi-condition environment adaptability experiments.

[0090] Based on the width of the first-level dynamic threshold range, the shrinkage range is calculated in combination with the link adaptation correction coefficient. The smaller the link adaptation correction coefficient, the stronger the constraint of the link characteristics on the threshold, and the larger the threshold shrinkage range. The larger the adjustment range, the weaker the link constraint, and the smaller the threshold shrinkage range, ensuring that the adjustment range is accurately adapted to the link characteristics and operating conditions.

[0091] To avoid excessive contraction in one direction causing the threshold to exceed the physical boundaries of device operation, the calculated contraction amplitude is evenly distributed across the upper and lower boundaries of the first-level dynamic threshold. Specifically, the upper limit of the first-level dynamic threshold is contracted downward by half of this amplitude, and the lower limit is contracted upward by half of this amplitude. This balanced contraction forms the final range of the second-level dynamic threshold, which enhances the sensitive identification of link anomalies while ensuring the rationality and physical feasibility of the threshold.

[0092] Specifically, the link adaptation correction coefficient library is a structured database storing the baseline correction coefficients and dynamic adjustment rules corresponding to different core collaborative links. Its core function is to provide a quantitative basis for the secondary constraints of the first-level dynamic threshold range. By combining the inherent characteristics of the core collaborative links, it optimizes the threshold boundaries, improves the adaptability of the second-level dynamic thresholds to the actual operating state of the links, and reduces fault misjudgments caused by differences in link type and working characteristics. Configuration is performed using a combination of offline calibration and online dynamic correction. For example, the baseline correction coefficient for the hydraulic main oil circuit link is 0.92, suitable for compression mode and compression-shear hybrid mode.

[0093] The primary and secondary dynamic threshold ranges of each monitoring parameter are linked and integrated, and the fault-sensitive point association identifiers and monitoring priority identifiers corresponding to each parameter are added. The threshold system verification process is executed by calling the typical working condition dataset, comparing the verification results with the preset verification standards, and iteratively correcting the threshold ranges that do not meet the standards. After the verification is passed, a complete fault judgment threshold system is output.

[0094] The benchmark construction module is used to collect real-time monitoring parameters of the dynamic compression-shear testing machine divided by loading cycles, add cycle sequence number identifiers to the monitoring data of each round, and bind them with the dynamic physical parameters of the core collaborative link to generate a time-series monitoring dataset; configure the benchmark candidate range, and call the first-level dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system generated by the parameter configuration module as the basis for judging the operating status of each loading cycle within the benchmark candidate range; compare the matching degree of each candidate cycle monitoring data with the corresponding first-level dynamic threshold range for each monitoring parameter; if all monitoring parameters of a loading cycle are within the corresponding first-level dynamic threshold range, it is judged as a normal operating state; otherwise, it is judged as an abnormal operating state; when the dynamic compression-shear testing machine of each loading cycle within the benchmark candidate range is in a normal operating state, the fault judgment benchmark establishment operation is triggered. By extracting the physical parameters and monitoring parameter feature values ​​of each core collaborative link through the benchmark feature extraction process, the fault judgment benchmark library corresponding to this motion mode is constructed, and the associated motion mode, core collaborative link identifier, and benchmark construction timestamp are fixed. If any abnormal loading loop exists within the candidate benchmark range, a benchmark construction error message will be output directly and the process will be terminated. At the same time, fault judgment benchmark information will be output to provide a basis for comparison for subsequent modules, and real-time consistency verification based on the benchmark will be supported. If the verification fails, the current benchmark will be terminated and a prompt to rebuild will be made to ensure that the benchmark is always based on the initial running state of the device in full normal operation, thus ensuring the accuracy of subsequent fault judgment.

[0095] In this embodiment, the benchmark construction module addresses the shortcomings of existing technologies that rely solely on preset fault judgment threshold systems for fault identification. These technologies lack a real-time initial operating benchmark that precisely matches the current motion mode of the dynamic compression-shear testing machine, making it difficult to accurately define the normal initial state of the current operation. Furthermore, they cannot capture minute changes in the monitored parameters during each loading cycle, and can only passively determine faults after the parameters exceed the threshold, failing to provide early compensation and correction. At the same time, it is difficult to ensure the adaptability of subsequent predictive compensation to the current working conditions, ultimately leading to delayed fault warnings and insufficient compensation accuracy. Therefore, by constructing a dedicated fault judgment benchmark library, the system can achieve accurate identification of the initial state and timely response to minute changes, thereby improving the monitoring, early warning, and compensation correction capabilities of the entire system.

[0096] Preferably, the specific steps for generating a time-series monitoring dataset include:

[0097] The system receives parameter configuration information from the parameter configuration module that matches the current motion mode. Based on the monitoring parameter type, it deploys corresponding signal acquisition units for each fault-sensitive point in the core collaborative link, starts the real-time data acquisition process at the preset acquisition frequency, and obtains the monitoring parameter data during the loading cycle of the dynamic compression-shear testing machine.

[0098] The core collaborative link dynamic physical parameter acquisition unit is invoked to acquire the dynamic physical parameters of the core collaborative link in real time. The dynamic physical parameters refer to the characteristic parameters that are strongly coupled with the operating state of the core collaborative link, change in real time with the loading cycle of the dynamic compression and shear testing machine, and can characterize the dynamic working characteristics and parameter transmission law of the hydraulic sub-link, mechanical sub-link, and control sub-link. These include the hydraulic circuit pressure transmission coefficient, the real-time gap value of mechanical components, and the signal response hysteresis coefficient of the electrical control system.

[0099] Based on the loading cycle control signal of the dynamic compression-shear testing machine, the start and end nodes of each loading cycle are identified. Using this as the basis, the continuously collected monitoring parameter data and dynamic physical parameters are segmented according to the loading cycle, forming segmented data groups divided by loading cycle.

[0100] According to the loading cycle sequence, a unique loading cycle number identifier is assigned to each segment of data group. The loading cycle number identifier includes the cycle sequence code and the cycle start and end timestamps.

[0101] The monitoring parameter segment data and dynamic physical parameter segment data under the same loading cycle sequence number are bound one-to-one to ensure the temporal consistency and link correlation of the two types of data. A validity verification process is then initiated, verifying data integrity, temporal consistency, and parameter correlation. Segment data groups with missing data, disordered timing, or invalid correlations are removed. The verified segment data groups are integrated to generate a time-series monitoring dataset containing the loading cycle sequence number, monitoring parameters, and dynamic physical parameters, and stored in the real-time database to provide data support for subsequent baseline candidate range selection and operational status assessment.

[0102] Please see Figure 2 Preferably, the specific steps for constructing the fault judgment benchmark library corresponding to this motion mode include:

[0103] Configure the baseline candidate range. The baseline candidate range refers to the initial loading cycle interval defined for constructing the fault judgment baseline for this motion mode. It is the range of valid data sources selected from the time-series monitoring dataset for baseline construction. Its boundaries are determined by the preset number of candidate cycles and the loading cycle timing sequence, and are used to ensure that the baseline is based on stable normal operating state data during the equipment startup phase. For example, the baseline candidate range is set to the first N loading cycles in the time-series monitoring dataset, where N is a preset positive integer. Based on the loading stability characteristics of the dynamic compression-shear testing machine and verified offline, all loading cycle segment data groups within the baseline candidate range in the time-series monitoring dataset are extracted and integrated to form the baseline candidate dataset.

[0104] The fault judgment threshold system generated by the parameter configuration module is invoked, and the first-level dynamic threshold range corresponding to each monitoring parameter is extracted as the basis for judging the running status. The full normality detection process of the candidate loop is executed. The matching degree between the monitoring parameters of each segment of the benchmark candidate dataset and the corresponding first-level dynamic threshold range is compared for each loading loop and each monitoring parameter. If all monitoring parameters of a single loading loop are within the corresponding first-level dynamic threshold range, the loading loop is determined to be a normal loading loop; otherwise, it is determined to be an abnormal loading loop.

[0105] If all loading loops within the candidate benchmark range are determined to be normal loading loops, the candidate benchmark dataset is confirmed to be valid, and the fault judgment benchmark establishment operation is triggered; if any abnormal loading loop exists, a benchmark construction anomaly prompt is directly output, the abnormal loading loop number and the corresponding anomaly monitoring parameters are recorded synchronously, and the current fault judgment benchmark library construction process is terminated.

[0106] Based on the benchmark candidate dataset, feature values ​​of each core collaborative link are extracted. These feature values ​​include monitoring parameter feature values ​​and dynamic physical parameter feature values. Simultaneously, the inherent physical parameters of the core collaborative links are used as supplementary static attributes of the benchmark. These inherent physical parameters are determined by the hardware structure and design parameters of the core collaborative links of the dynamic compression-shear testing machine and are inherent characteristic parameters that do not dynamically change with the loading cycle. They are used to characterize the basic working characteristics of the hydraulic sub-link, mechanical sub-link, and control sub-link, providing static attribute support for the benchmark features and ensuring that the benchmark can fully reflect the inherent operating laws of the links, such as the inherent stiffness of the hydraulic circuit, the stiffness of mechanical components, and the basic adjustment gain of the electronic control system. The extracted feature values ​​include the temporal variation law, peak value characteristics, valley value characteristics, and rate of change characteristics of parameters within a single loading cycle, as well as the consistency fluctuation range characteristics of parameters between different loading cycles.

[0107] The feature dimensions are divided according to the core collaborative links. The feature values ​​of monitoring parameters and dynamic physical parameters under the same link are associated and bound with the inherent physical parameters to generate a baseline feature subset corresponding to each core collaborative link. Each baseline feature subset must clearly include parameter feature information, the correlation mapping relationship between parameters, and the link affiliation identifier.

[0108] Integrate the subset of benchmark features of all core collaborative links to form a fault judgment benchmark library corresponding to this motion mode; supplement and solidify the associated information, including the identifier of this motion mode, the identifier of the core collaborative link, the benchmark construction timestamp, the benchmark candidate range information and the corresponding parameter configuration version number, to ensure the uniqueness and traceability of the benchmark library.

[0109] The system invokes preset benchmark feature verification rules, which verify the completeness, consistency, and relevance of benchmark features. If the verification passes, the complete fault judgment benchmark library is stored in the benchmark database. If the verification fails, the system returns to re-execute the benchmark feature extraction process. If the extraction fails three times in a row, a benchmark construction error message is output and the process is terminated.

[0110] The prediction and compensation module is used to acquire segmented data sets of subsequent loading cycles of the dynamic compression-shear testing machine based on the time-series monitoring dataset, extract the monitoring parameters and dynamic physical parameters of the current loading cycle, perform deviation analysis based on the fault judgment benchmark library, call the benchmark feature subset of the corresponding core collaborative link in the fault judgment benchmark library, execute the deviation calculation process between real-time data and benchmark features, construct a deviation trend prediction model based on the deviation data of historical multiple loading cycles, predict the deviation trend, and generate the deviation prediction value and confidence interval of the monitoring parameters in the next loading cycle. It calls the secondary dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system generated by the parameter configuration module for constraint verification to determine whether the compensation strategy generation process is triggered. If so, it matches the preset compensation strategy library according to the deviation trend type and deviation magnitude, generates and executes a targeted compensation strategy to achieve early correction of operational deviations. It also collects monitoring data of the loading cycle after compensation in real time to verify the compensation effect.

[0111] In this embodiment, the core of the predictive compensation module lies in addressing the shortcomings of existing technologies, such as the lack of a dynamic deviation prediction mechanism based on real-time operating conditions, the poor adaptability of compensation strategies to the current motion mode, and the inability to predict deviation trends in advance, leading to compensation lag. By combining deviation analysis with a fault judgment benchmark library specific to this work, and using a two-level dynamic threshold to define the compensation boundary, the module ensures the pertinence and effectiveness of the compensation strategy. Simultaneously, it achieves early compensation through deviation trend prediction, preventing deviation accumulation from evolving into a fault, further improving the operational stability and testing accuracy of the dynamic compression-shear testing machine. The compensation strategy library refers to a database storing the mapping relationship between deviation trend types, deviation magnitudes, and corresponding compensation instructions. Its mapping rules are derived from the physical characteristics of the core collaborative link, historical compensation cases, and offline verification results.

[0112] Please see Figure 3 Preferably, the specific steps for predicting the deviation trend and generating the predicted deviation values ​​and confidence intervals of the monitored parameters in the subsequent loading cycle include:

[0113] The system continuously receives segmented data groups from subsequent loading loops in the time-series monitoring dataset; it calls the primary dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system, and compares the matching degree of the monitoring parameters of the current loading loop segmented data group with the primary dynamic threshold for each monitoring parameter; if all monitoring parameters are within the primary dynamic threshold range, the current loading loop is determined to be a normal loading loop, and deviation analysis is triggered; if any monitoring parameter exceeds the primary dynamic threshold range, it is determined to be an abnormal loading loop, the deviation prediction process is terminated, and it is directly pushed to the fault identification module.

[0114] Extract monitoring parameters and dynamic physical parameters for cycles deemed to be in normal loading. Simultaneously, backtrack the historical normal loading cycle segment data groups from the preset monitoring period prior to the current loading cycle to extract corresponding historical monitoring parameters and historical dynamic physical parameters. The preset monitoring period refers to the time window range pre-set for backtracking historical normal loading cycle data to ensure the accuracy of deviation trend prediction. Specifically, it consists of M consecutive normal loading cycles, where M is a preset positive integer, set after offline verification based on loading stability under different motion modes and the input requirements of the prediction model. This provides a sufficient number of historical deviation data samples with temporal continuity for the deviation trend prediction model, ensuring that the model can fully learn the evolution law of deviation and improve the credibility and accuracy of the prediction results.

[0115] Perform smoothing filtering preprocessing and data alignment operations on the segmented data groups of the current loading loop and historical normal loading loops, and integrate them to form a dataset for predictive analysis, ensuring the temporal continuity and parameter integrity of the dataset.

[0116] The system calls upon the fault diagnosis benchmark library and matches the corresponding benchmark feature subsets according to the core collaborative link dimension. Based on the dataset used for predictive analysis, deviation data is calculated parameter by parameter and link by link. The deviation data includes instantaneous deviation, cumulative deviation within a cycle, and deviation change rate. A deviation analysis matrix is ​​formed, containing the core collaborative link identifier, parameter identifier, instantaneous deviation, cumulative deviation within a cycle, and deviation change rate. Instantaneous deviation refers to the degree of deviation between the actual measured value of a certain monitoring parameter or dynamic physical parameter at a specific time node in the current loading cycle and the benchmark value of the same parameter at the same time node in the corresponding core collaborative link benchmark feature subset in the fault diagnosis benchmark library. This is obtained by calculating the difference between the parameter value at each time node in the current loading cycle and the corresponding node value of the benchmark feature. The cumulative deviation within a cycle is obtained by integrating all instantaneous deviations in the current loading cycle to obtain the cumulative deviation value within a single cycle. The deviation change rate is obtained by calculating the difference between the instantaneous deviations of two adjacent time nodes.

[0117] Deviation data from the current loading cycle and historical normal loading cycles are extracted from the deviation analysis matrix and arranged in chronological order to form a historical deviation time series dataset. Abnormal abrupt changes in the historical deviation time series dataset are removed to ensure that the historical deviation time series dataset meets the input requirements of the time series prediction model and to avoid abnormal data interfering with the discovery of prediction patterns.

[0118] The system invokes a pre-built deviation trend prediction model and automatically loads matching model parameters based on the current motion pattern. It then substitutes the processed historical deviation time-series dataset into the deviation trend prediction model, outputting the predicted deviation values ​​and confidence intervals for each parameter in the subsequent loading loop. Simultaneously, it generates a deviation trend curve to visually represent the direction of deviation evolution. The deviation prediction results and trend curve are then pushed to the secondary dynamic threshold constraint verification stage.

[0119] The deviation trend prediction model consists of five functional layers: a data input layer, a bidirectional LSTM feature extraction layer, an attention weight allocation layer, a fully connected prediction layer, and an output layer. The data input layer receives historical deviation time-series datasets with an input dimension of [T, F]. T represents the number of historical normal loading loops, and F represents the deviation feature dimension of a single loop, including three core features: instantaneous deviation, cumulative deviation within the loop, and deviation change rate, along with two auxiliary features: core collaborative link identifier and parameter type. Before input, the data needs to be normalized to eliminate dimensional differences. A bidirectional LSTM feature extraction layer is used to mine the bidirectional temporal correlation features of deviation data in the historical deviation time-series dataset, fully capturing the forward and backward dependencies of deviation evolution with loading cycles, and obtaining a deviation feature matrix. A two-layer bidirectional LSTM network is set, with 64 neurons in each hidden layer. The forward LSTM learns the temporal progression of deviation data, such as the trend of deviation increasing with the number of cycles, while the backward LSTM learns the reverse correlation characteristics of deviation data, such as the dependency of the current deviation on historical deviations. By concatenating the feature vectors from the forward and backward outputs, a deviation feature matrix containing complete temporal information is obtained. An attention weight allocation layer is used to differentiate the weights of the deviation feature matrix output by the bidirectional LSTM, obtaining a weighted feature vector that highlights key temporal node features that significantly influence deviation trend prediction and suppresses interference from irrelevant features. A soft attention mechanism is introduced to calculate the contribution weight of each historical loading cycle deviation feature to the current prediction task. The deviation feature matrix output by the bidirectional LSTM is weighted and summed to focus on key features that significantly affect the deviation trend, improving the model's predictive accuracy. The fully connected prediction layer performs non-linear transformation and dimensionality mapping on the weighted feature vector output from the attention weight allocation layer, achieving accurate conversion from historical bias features to bias prediction values ​​in subsequent loading loops. A two-layer fully connected network is used: the first layer has 32 neurons with ReLU activation, and the second layer has 1 neuron with linear activation. This maps the weighted feature vector output from the attention layer to the bias prediction values ​​in subsequent loading loops. The output layer integrates the prediction results from the fully connected layers, outputting bias prediction information with confidence support. Based on the prediction results from the fully connected layers and combined with the statistical distribution of prediction errors during model training, the 95% confidence interval of the bias prediction values ​​is calculated and output. Simultaneously, the feature contribution ranking is output, clearly identifying the core parameters influencing the bias trend.

[0120] Preferably, the specific steps for generating and implementing a targeted compensation strategy include:

[0121] Based on the values ​​of each monitoring parameter in the current loading cycle and the predicted deviation values ​​and confidence intervals of the monitoring parameters in the next loading cycle, the predicted values ​​and confidence intervals of each monitoring parameter in the next loading cycle are generated; the dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system is called, and the deviation status is determined for each monitoring parameter. The deviation status includes no compensation requirement, ordinary compensable deviation, critical compensable deviation and uncompensable fault deviation.

[0122] Specifically, if the confidence interval of each monitoring parameter in the subsequent loading cycle is within the range of the first-level dynamic threshold, it is determined that there is no need for compensation and the system returns to continuous monitoring. Otherwise, it is determined whether the predicted value of each monitoring parameter in the subsequent loading cycle is within the range of the second-level dynamic threshold. If so, it indicates that the monitoring parameter deviation has a potential risk of exceeding the second-level dynamic threshold, but the current predicted value is still within the controllable range of the second-level dynamic threshold, and is determined to be a normal compensable deviation. If not, it is determined whether the predicted value of each monitoring parameter in the subsequent loading cycle is within the range of the first-level dynamic threshold. If so, it indicates that the parameter deviation has exceeded the constraint of the second-level dynamic threshold and is within the critical risk range of the first-level dynamic threshold, and is determined to be a critical compensable deviation, triggering the enhanced compensation mechanism. If not, it indicates that the parameter deviation has exceeded the constraint of the first-level dynamic threshold and exceeds the system's autonomous compensation capability, and is determined to be an uncompensable fault deviation. The compensation process is directly terminated and a fault prediction and early warning are issued, and early warning information is output synchronously, including the location of the fault sensitive point, the name and value of the deviation parameter, and the magnitude of the deviation exceeding the first-level dynamic threshold. This provides data support for accurate identification of fault types and location of fault causes.

[0123] For monitoring parameters classified as ordinary compensable deviations or critically compensable deviations, multi-dimensional attribute information is extracted based on the deviation analysis matrix, specifically including: link affiliation attribute, deviation magnitude attribute, and trend evolution attribute. These are integrated into standardized deviation attribute labels to provide a basis for accurate matching of compensation strategies. Among them, the link affiliation attribute clarifies the core collaborative sub-links corresponding to the deviation, including hydraulic sub-links, mechanical sub-links, and electrical control sub-links, as well as specific fault-sensitive points. The deviation magnitude attribute is divided into slight, moderate, and severe deviation levels according to preset proportions, such as 5% and 15% of the width of the secondary dynamic threshold interval. The trend evolution attribute is used to indicate the rate of change and direction of change of the deviation, including increasing, decreasing, and periodic fluctuations.

[0124] The system calls upon a pre-defined compensation strategy library and executes a compensation strategy matching process based on deviation attribute labels. The library contains built-in three-dimensional mapping rules for link affiliation attributes, deviation magnitude attributes, and trend evolution attributes. These rules are used to match compensation strategies based on multi-dimensional attribute information. The mapping rules are generated through core collaborative link physical characteristic analysis, historical compensation case training, and offline working condition verification optimization. For example, for ordinary compensable deviations, a basic compensation strategy is matched, with the strategy content being single-dimensional parameter adjustment. For instance, slight pressure deviations in hydraulic sub-links use pressure closed-loop fine-tuning commands, and slight clearance deviations in mechanical sub-links use clearance adaptive compensation commands. For critical compensable deviations, an enhanced compensation strategy is matched, superimposed with a dual enhancement mechanism on top of the basic compensation strategy: first, increasing the adjustment amplitude of the compensation parameter by 10%-20% according to a preset ratio; second, shortening the compensation execution cycle from once per cycle to twice per cycle, executed before the loading cycle starts and at the loading peak node, respectively.

[0125] The matched compensation strategy is transformed into executable standardized compensation instructions. These instructions include: target actuator identifier, compensation control parameters, execution timing constraints, and safety threshold constraints. The target actuator identifier identifies the recipient of the instruction, such as a hydraulic valve group, mechanical adjustment mechanism, or electronic control unit. Compensation control parameters include specific values ​​such as adjustment amplitude, adjustment frequency, and duration. Execution timing constraints mark the instruction trigger points. Safety threshold constraints set the upper limit for the adjustment of compensation parameters to prevent new parameter exceedance risks caused by over-compensation. A unique identifier is added to each compensation instruction for easy traceability throughout the entire process.

[0126] Standardized compensation instructions are issued to the corresponding actuators. After receiving the instructions, the actuators perform compensation operations synchronously with the next loading cycle according to the preset timing constraints: the hydraulic sub-link performs pressure and flow regulation, the mechanical sub-link performs clearance compensation and stiffness adjustment, and the electrical control sub-link performs voltage, current, and gain correction. During the execution process, the actuators' motion status data are collected in real time to ensure that the compensation operations are accurately implemented.

[0127] After the next loading cycle is completed, the compensated monitoring parameters are extracted. If the compensated monitoring parameters are within the range of the secondary dynamic threshold, the compensation is deemed effective. If the deviation of the compensated monitoring parameters is outside the range of the secondary dynamic threshold, a targeted compensation strategy is regenerated and executed. The number of times the targeted compensation strategy is regenerated and executed is counted. If it exceeds the preset control threshold, the compensation is deemed ineffective, and a compensation anomaly warning is issued. The preset control threshold refers to the maximum number of times the compensation strategy can be re-executed, pre-set to avoid fluctuations in equipment operating status due to over-compensation and to prevent waste of system resources due to invalid compensation retries. This threshold is determined based on the compensation response characteristics of each core collaborative link of the dynamic compression-shear testing machine, the statistics of historical compensation failure cases, and the offline working condition verification results. An example value of 3 times is used to define the effective retry boundary of the compensation strategy. By limiting the number of retries, the compensation accuracy and equipment operating stability are balanced. When the number of retries exceeds this threshold, it is clearly determined that the current deviation has exceeded the system's normal compensation capability. The compensation process is terminated in a timely manner, and an anomaly warning is initiated to ensure the timeliness of subsequent fault handling.

[0128] The fault identification module is used to identify potential fault monitoring parameters based on the above-mentioned fault judgment threshold system, according to the actual values ​​of each monitoring parameter in the current loading cycle and the predicted values ​​of each monitoring parameter in the subsequent loading cycle. It identifies potential fault monitoring parameters by comparing and analyzing the coupling judgment rules of the actual value exceeding the limit and the predicted value exceeding the limit trend, and triggers the fault identification operation. It obtains link fault feature evidence by extracting physical features from each potential fault monitoring parameter, constructs a link-based fault causal graph, calculates the fault probability of each core collaborative link through causal reasoning, identifies faulty sub-links, and performs fault type diagnosis. It accurately distinguishes the main fault link and the secondary fault link in the compound fault scenario, clarifies the fault propagation path, and determines the fault type by matching the association mapping relationship between link fault feature evidence and fault type in the fault knowledge base.

[0129] In this embodiment, the core of the fault identification module is to solve the problems of single fault identification trigger logic, difficulty in distinguishing the primary and secondary relationships of compound faults, fuzzy fault link location, and poor adaptability to the multiple motion modes of dynamic compression and shear testing machines in the prior art. The prior art usually judges the fault only based on whether the actual value of the monitored parameter exceeds the threshold, ignoring the pre-fault warning role of the parameter evolution trend, which easily leads to delayed fault identification or misjudgment. At the same time, in the face of compound fault scenarios with multiple abnormal links, it is impossible to define the root cause of the primary fault and the derivative relationship of the secondary fault, resulting in low efficiency of operation and maintenance. The fault identification module achieves precise pre-fault triggering by coupling the actual value exceeding the limit state with the predicted value exceeding the limit trend. It extracts the physical characteristics of potential fault monitoring parameters to form link fault characteristic evidence, and constructs a link-based fault causal graph based on the power transmission characteristics of the core collaborative links, visualizing the fault propagation path. Based on causal reasoning algorithms, it calculates the fault probability of each core collaborative link, and combines historical cases and mapping rules from the fault knowledge base. This not only accurately locates faulty sub-links but also distinguishes between primary and secondary fault links in compound faults, clarifying the root cause of the fault and the causal relationship of derived faults, thus addressing the industry pain point of compound fault diagnosis that only treats the symptoms, not the root cause. The fault knowledge base is a structured database storing core collaborative link fault characteristic evidence, fault types, fault causal relationships, and historical fault handling cases. For example, the hydraulic valve jamming fault characteristics of the hydraulic sub-link are pressure fluctuation exceeding 5% and response delay exceeding 80ms, and the handling is cleaning the valve core and calibrating it. Its data sources include offline fault simulation experimental data from different motion modes of dynamic compression and shear testing machines, on-site operation and maintenance fault records, and industry standard fault diagnosis specifications, supporting online dynamic updates based on new fault cases.

[0130] Preferably, the specific steps for identifying potential fault monitoring parameters include:

[0131] The system receives the actual value dataset of each monitoring parameter in the current loading loop, as well as the predicted values ​​and confidence intervals of each monitoring parameter in the next loading loop. Simultaneously, it invokes the fault judgment threshold system corresponding to the current motion mode, extracts the primary and secondary dynamic threshold ranges for each monitoring parameter, and associates the link attribution identifier and fault-sensitive point location information for each parameter. The system then binds and associates the above data according to the structure of monitoring parameter ID, link attribution identifier, fault-sensitive point location, numerical type, and dynamic threshold range to generate a judgment dataset, ensuring the link correlation and threshold matching of the data. The numerical type includes actual values ​​and predicted values.

[0132] Based on the link attribution identifier of the judgment dataset, the monitoring parameters are divided into multiple core collaborative sub-link parameter groups according to the core collaborative link, including hydraulic sub-link parameter group, mechanical sub-link parameter group, and electrical control sub-link parameter group. Within each core collaborative sub-link parameter group, the parameter attributes of the corresponding fault sensitive points are further associated. For example, the pressure parameter of the hydraulic sub-link is associated with the fault sensitive point of the hydraulic valve group, and the flow parameter is associated with the fault sensitive point of the pipeline. This lays the foundation for accurate determination of potential fault sources in the future.

[0133] For the monitoring parameters of each core collaborative sub-link parameter group after splitting, a coupled determination of the actual value exceeding the limit and the predicted value exceeding the limit trend is performed for each monitoring parameter. The specific determination rules are as follows:

[0134] If the actual value of a monitoring parameter in the current loading loop exceeds the corresponding first-level dynamic threshold range, the parameter is directly determined to be a potential fault monitoring parameter, and the over-limit type is marked as actual value over-limit. At the same time, the over-limit magnitude and the over-limit duration are recorded. If the actual value exceeds the limit for multiple consecutive rounds, the duration is accumulated. The over-limit magnitude refers to the quantification degree to which the actual value of the monitoring parameter exceeds the boundary of the corresponding first-level dynamic threshold range in the current loading loop, which is used to accurately characterize the severity of parameter anomalies. Its specific determination logic is as follows: First, determine whether the actual value of the monitoring parameter breaks through the upper or lower boundary of the first-level dynamic threshold. Then, calculate the absolute difference between the actual value and the boundary. Finally, divide the absolute difference by the interval width between the upper and lower limits of the first-level dynamic threshold. The result is the over-limit magnitude.

[0135] If the actual value of a monitoring parameter in the current loading cycle does not exceed the first-level dynamic threshold range, but the predicted value in a subsequent loading cycle exceeds the first-level dynamic threshold range, or the confidence interval intersects with the first-level dynamic threshold range, then trend tracing judgment is initiated:

[0136] Extract the historical actual value data of the monitoring parameter in the historical loading cycle and calculate the time-series change slope; if the absolute value of the time-series change slope of the historical actual value is greater than the preset trend threshold, the preset trend threshold is set offline based on the physical characteristics of different parameters, and the direction of the predicted value exceeding the limit is consistent with the trend of the time-series change slope, then the monitoring parameter is determined to be a potential fault monitoring parameter, and the exceedance type is marked as predicted value exceedance; otherwise, it is determined to be a suspected exceedance parameter, and is not included in the list of potential fault monitoring parameters for the time being. The monitoring status is returned to continuous monitoring and the next round of actual value of the parameter is collected in encrypted form.

[0137] A list of potential fault monitoring parameters is constructed based on the monitoring parameter ID, link affiliation identifier, fault sensitive point location, over-limit type, and over-limit magnitude of the potential fault monitoring parameters.

[0138] Please see Figure 4 Preferably, the specific steps for identifying faulty sub-links and performing fault type diagnosis include:

[0139] Based on the established list of potential fault monitoring parameters, physical features are extracted for each parameter to form a link fault feature evidence set. The extracted physical features include core anomaly features and link-related features. Core anomaly features are a set of quantitative features that directly reflect the degree of anomaly, evolution pattern, and risk level of potential fault monitoring parameters. They are the core basis for judging the nature of parameter anomalies and fault development trends, specifically including key indicators such as over-limit type, over-limit amplitude level, over-limit duration, time-series change slope, and trend consistency. Link-related features are a set of features that reflect the relationship between potential fault monitoring parameters and core collaborative sub-links, fault-sensitive points, and other monitoring parameters within the same link. They are used to establish a mapping relationship between parameter anomalies and fault link attribution, clarifying the physical carrier and propagation associated objects of the fault. Specifically, they cover key information such as the attribution of the parameter to the corresponding core collaborative sub-link, the location of the fault-sensitive point, and the abnormal synergy of related parameters within the same link, such as the synchronous anomaly relationship between pressure and flow parameters in a hydraulic sub-link. The extracted physical features are structurally integrated according to the structure of core collaborative sub-links, fault-sensitive points, physical feature types, and physical feature quantification values ​​to generate feature evidence subsets corresponding to each core collaborative sub-link, ensuring a strong correlation between physical feature evidence and the link. Physical feature types refer to the classification results based on feature attributes and functions, specifically including two categories: abnormal core features and link-related features. These two types of features, from the perspectives of parameter anomaly characteristics and link correlation, together constitute a complete evidence chain for fault diagnosis.

[0140] By combining the power transmission logic of the core collaborative links, such as the hydraulic sub-link driving the mechanical sub-link, the control sub-link regulating the hydraulic system, and the mechanical sub-link operating, as well as the fault causal relationship data stored in the fault knowledge base, a link-based fault causal graph is constructed.

[0141] The nodes in the link-based fault cause-effect graph include core collaborative sub-link nodes, fault-sensitive point nodes, and core component nodes. Directed edges between nodes represent fault propagation relationships, and the edge weights represent propagation probabilities, set based on historical fault case statistics and industry fault evolution patterns. The specific construction process is as follows:

[0142] The fault-sensitive node corresponding to the potential fault monitoring parameter is used as the initial trigger node, and its core collaborative sub-link node is associated with it.

[0143] Based on the power transmission sequence of the core collaborative link, supplement the associated nodes of adjacent core collaborative sub-links. For example, associate the hydraulic sub-link nodes with the fault-sensitive nodes of the hydraulic valve group, and then associate them with the core component nodes such as hydraulic pumps and hydraulic cylinders.

[0144] By retrieving historical fault propagation data from the fault knowledge base, propagation probabilities are assigned to directed edges between nodes to complete the construction of a link-based fault causal graph, enabling visualized modeling of fault propagation paths.

[0145] The structured and integrated evidence set of link failure features is input into the constructed link-based fault causal graph. Causal inference algorithms, such as Bayesian network inference algorithms, are used to calculate the failure probability of each core collaborative sub-link. The specific inference process is as follows:

[0146] Using the physical characteristics of each potential fault monitoring parameter as input, the triggering conditions of the corresponding fault-sensitive node in the link-based fault causal graph are matched, and the node triggering probability is calculated. The triggering conditions of the fault-sensitive node refer to a set of pre-defined quantitative judgment rules based on the characteristic patterns of typical faults corresponding to the fault-sensitive point in the fault knowledge base, used to determine whether the sensitive point shows signs of fault. These conditions must be directly related to the physical characteristics of the potential fault monitoring parameters, and for example, include the following core judgment dimensions:

[0147] If the exceedance level of a potential fault monitoring parameter reaches or exceeds a preset level, the trigger condition for this dimension is met; if the cumulative duration of the exceedance state of a potential fault monitoring parameter reaches a preset duration, the trigger condition for this dimension is met; if the temporal change slope of a potential fault monitoring parameter is consistent with the characteristic trend of a typical fault at the fault-sensitive point, and the absolute value of the slope is greater than a preset trend threshold, the trigger condition for this dimension is met; if the monitoring parameter corresponding to the fault-sensitive point is abnormal, and related parameters within the same link simultaneously exhibit abnormal characteristics, the trigger condition for this dimension is met. Only when a preset number of conditions in the above dimensions are met (exemplary value: at least two conditions must be met; this preset number is set based on offline verification of historical fault diagnosis accuracy) can the fault-sensitive point node be determined to be triggered, thereby initiating the node trigger probability calculation process.

[0148] Based on the preset propagation probability of the directed edge between the fault-sensitive node and the core collaborative sub-link node, the comprehensive fault probability of a single path of the core collaborative sub-link corresponding to a single fault-sensitive node is recursively calculated along the fault propagation direction, and used as the contribution value of the potential fault monitoring parameter to the propagation probability of the corresponding core collaborative sub-link.

[0149] The final failure probability of each core collaborative sub-link is obtained by combining the node trigger probability and propagation probability contribution values ​​corresponding to all potential fault monitoring parameters within the same core collaborative sub-link. The value ranges from 0 to 1, with a higher value indicating a higher probability of failure in that core collaborative sub-link. The propagation probability contribution value refers to the quantitative contribution of the node trigger probability corresponding to a certain potential fault monitoring parameter within the same core collaborative sub-link, after weighting the fault propagation probabilities among nodes, to the final failure probability of that core collaborative sub-link. Its calculation logic is the product of the node trigger probability and the corresponding fault propagation probability. After accumulating the propagation probability contribution values ​​of multiple potential fault monitoring parameters, and combining them with the inherent fault susceptibility weight of the link, the final failure probability of the core collaborative sub-link is obtained.

[0150] Based on offline verification of historical fault diagnosis accuracy, a fault probability threshold is set, with an example value of 0.7. Core collaborative sub-links with a final fault probability higher than this fault probability threshold are identified as faulty sub-links, forming a list of faulty sub-links.

[0151] If the list of faulty sub-links is not unique, i.e., there are composite fault scenarios with multiple faulty sub-links, then the operation of distinguishing between primary and secondary faulty links is performed based on the fault propagation starting point, the time-series information of the potential fault monitoring parameters corresponding to the faulty sub-links, or the fault probability. For example, the fault propagation path in the causal reasoning process is traced, and the faulty sub-link corresponding to the fault propagation starting point is determined as a candidate primary faulty link; or the fault probabilities of each faulty sub-link are compared, and the faulty sub-link with the highest fault probability is selected as the candidate primary faulty link; or, combined with the time-series information of the potential fault monitoring parameters, if the abnormal time sequence of the potential fault monitoring parameters corresponding to a certain faulty sub-link is earlier than that of other faulty sub-links and is consistent with the fault propagation starting point, then it is finally determined as the core primary faulty link, and the remaining faulty sub-links are secondary faulty links, i.e., related faulty links derived from the primary fault. The fault propagation starting point refers to the core node that first fails in the composite fault scenario and serves as the source of subsequent fault propagation.

[0152] The structured feature evidence subset of each faulty sub-link is precisely matched with the fault type template in the fault knowledge base to perform fault type diagnosis. The specific matching process is as follows:

[0153] Candidate fault types in the fault knowledge base are filtered according to the fault sub-link affiliation. For example, hydraulic sub-links correspond to candidate types such as hydraulic valve sticking and pipeline leakage, while mechanical sub-links correspond to candidate types such as component wear and excessive clearance.

[0154] The matching degree between the subset of feature evidence and each candidate fault type template is calculated. Matching dimensions include consistency of exceedance features, conformity of link association features, and similarity of temporal evolution features. Consistency of exceedance features refers to the degree of agreement between the abnormal core features of the current potential fault monitoring parameters and the typical exceedance features defined by the candidate fault type templates in the fault knowledge base. Matching requires comparing the consistency of indicators such as exceedance type, exceedance magnitude level, and exceedance duration. Conformity of link association features refers to the degree of agreement between the link association features of the current potential fault and the link affiliation, fault sensitive point location, and synergy of the same link parameters required by the candidate fault type templates in the fault knowledge base. Similarity of temporal evolution features refers to the degree of similarity between the dynamic evolution features of the current potential fault monitoring parameters, such as the temporal change slope and trend consistency, and the typical temporal evolution features recorded by the candidate fault type templates in the fault knowledge base. Matching requires comparing the similarity of abnormal trends in the parameters, including indicators such as increasing, decreasing, periodic fluctuations, and temporal change slope intervals. The relevant preset thresholds are the core quantitative standards for determining the triggering of fault-sensitive nodes. They are used to accurately distinguish between normal parameter fluctuations and fault signs. Specifically, they include: preset levels for the severity of exceeding limits; preset durations for the duration of exceeding limits to distinguish between occasional fluctuations and continuous anomalies; preset trend thresholds for the slope of time-series changes to determine whether parameter changes are a fault development trend; thresholds for determining whether parameters on the same link are abnormally synchronized; and preset thresholds for the number of trigger conditions to be met to balance diagnostic accuracy and false negative rate and determine the minimum number of conditions required for node triggering. Each threshold is configured in conjunction with historical fault patterns, monitoring parameters, link characteristics, and offline verification to ensure the accuracy of node triggering determination.

[0155] The candidate type with the highest matching degree and above the preset matching threshold is selected as the final fault type. If multiple types with similar matching degrees exist, further verification is performed using historical fault handling cases and core component model parameters from the fault knowledge base to determine a unique fault type. For primary and secondary fault links, fault type diagnosis is completed separately to clarify the causal relationship between the primary and secondary fault types. The preset matching threshold is a critical value used to determine the effectiveness of the matching between the subset of link fault feature evidence and the candidate fault type templates in the fault knowledge base. It is configured through a combination of offline statistical analysis and multi-condition verification.

[0156] Working principle and its effects:

[0157] This invention uses dynamic adaptation, benchmark anchoring, advance compensation, and precise diagnosis as its core logic to achieve full-process monitoring and fault early warning of the operating status of a dynamic compression-shear testing machine. It not only solves the pain points of poor adaptability, delayed early warning, and ambiguous fault location of traditional monitoring solutions, but also greatly improves the stability of equipment operation and the reliability of test data.

[0158] After the testing machine is started, it first matches the monitoring parameters corresponding to the motion mode with the core collaborative link to generate primary and secondary dynamic thresholds to construct a fault judgment system. This breaks through the limitations of traditional fixed thresholds and reduces the risk of false positives and false negatives. Then, it generates a time-series dataset by collecting data in a loading loop, and filters normal data to build a benchmark library specifically for operating conditions, avoiding the omission of minor deviations. Subsequently, based on the benchmark library, it predicts the trend of subsequent parameter deviations and combines it with a secondary threshold matching compensation strategy to achieve early control and suppress the accumulation of deviations into faults. Finally, it integrates the actual and predicted values ​​of parameters to identify potential faults, locates faulty sub-links through causal reasoning, distinguishes between primary and secondary links, and diagnoses fault types, shortening maintenance time and reducing costs.

[0159] In summary, this invention, through the progressive and synergistic empowerment of each module, integrates dynamic adaptation throughout the entire process of parameter configuration and threshold generation. It establishes a solid foundation for accurate deviation analysis and fault diagnosis through benchmark anchoring, achieves proactive fault prevention through advance compensation, and improves fault handling efficiency through precise diagnosis, forming a closed-loop management system from parameter adaptation to fault handling. This not only significantly improves the accuracy of dynamic compression-shear testing machine operation status monitoring and the foresight of fault early warning, but also effectively ensures long-term stable operation of the equipment through precise fault root cause location and advance compensation, improves the reliability of test data, reduces equipment maintenance costs, and provides an efficient solution for the intelligent operation and maintenance of dynamic compression-shear testing machines.

[0160] The above description is merely a preferred embodiment of the present invention, and the scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principle of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. An online monitoring and fault early warning system for the running state of a dynamic compression-shear testing machine, characterized in that, The parameter configuration module is configured to match monitoring parameters of a core collaborative link fault sensitive point and generate a dynamic threshold range of each monitoring parameter according to a current motion mode, and to construct a fault judgment threshold system. The reference construction module is configured to collect real-time monitoring parameters and dynamic physical parameters of a dynamic compression-shear testing machine divided according to a loading cycle, generate a time series monitoring data set, configure a reference candidate range, and judge whether the dynamic compression-shear testing machine of each loading cycle in the reference candidate range is in a normal operating state based on the fault judgment threshold system, if so, trigger a fault judgment reference establishment operation, and construct a fault judgment reference library. The prediction compensation module is configured to perform deviation analysis on the time series monitoring data set based on the fault judgment reference library, predict a deviation trend, generate predicted values and a confidence interval of each monitoring parameter in a subsequent round of loading cycle, and judge whether to trigger a compensation strategy generation process, if so, generate a compensation instruction according to a deviation trend type and a deviation magnitude. The fault identification module is configured to identify potential fault monitoring parameters according to actual values of each monitoring parameter in the current loading cycle and predicted values of each monitoring parameter in the subsequent round of loading cycle based on the fault judgment threshold system, and construct a potential fault monitoring parameter list. The fault identification module is configured to identify potential fault monitoring parameters according to actual values of each monitoring parameter in the current loading cycle and predicted values of each monitoring parameter in the subsequent round of loading cycle based on the fault judgment threshold system, and construct a potential fault monitoring parameter list. The step of identifying a fault sub-link and performing fault type diagnosis includes: The physical features include abnormal core features and link associated features. The extracted physical features are structured and integrated according to the structure of core collaborative sub-links, fault sensitive points, physical feature types, and physical feature quantitative values, to generate feature evidence subsets corresponding to each core collaborative sub-link. The link-based fault causal graph is constructed in combination with a power transmission logic of the core collaborative link and fault causal relationship data stored in a fault knowledge base. The nodes of the link-based fault causal graph include core collaborative sub-link nodes, fault sensitive point nodes, and core component nodes. The link fault feature evidence set is input into the constructed link-based fault causal graph, and a causal reasoning algorithm is used to calculate the final fault probability of each core collaborative sub-link. If the fault sub-link list contains more than one fault sub-link, a main fault link and a secondary fault link are distinguished according to a fault propagation starting point, time series information of a potential fault monitoring parameter corresponding to the fault sub-link, or a fault probability. ​ ​ The structured feature evidence subset of each fault sub-link is matched with the fault type templates in the fault knowledge base, candidate fault types in the fault knowledge base are screened according to the fault sub-link attribution, and the matching degrees of the feature evidence subset and each candidate fault type template are calculated to identify the final fault type.

2. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 1, characterized in that, The dynamic threshold range comprises a primary dynamic threshold range and a secondary dynamic threshold range; The step of constructing the fault judgment threshold system comprises: Before executing the test task, the motion mode of the current dynamic compression and shear testing machine is acquired, and a core cooperative link matching the current motion mode is selected; The core cooperative link is analyzed, and a fault sensitive point of the current core cooperative link is extracted; the fault sensitive point is a key link of the core cooperative link that needs to be monitored for faults; The monitoring parameters of the fault sensitive point of the core cooperative link are acquired, a monitoring parameter list is generated, and corresponding standard threshold ranges are configured for the monitoring parameters; The historical normal data of the current core cooperative link monitoring parameters are acquired, and are used to calculate distribution characteristic values of the monitoring parameters; the distribution characteristic values comprise a mean value, a standard deviation and a confidence interval; Based on the confidence interval in the distribution characteristic values, a reasonable fluctuation range of the monitoring parameters in the actual operation state of the current equipment is set; and according to the deviation amount between the reasonable fluctuation range and the standard threshold range, boundary correction of the standard threshold range is performed to acquire a primary dynamic threshold range of each monitoring parameter; A link adaptation correction coefficient of the current core cooperative link is matched, and the primary dynamic threshold range is constrained again to generate a secondary dynamic threshold range; The primary dynamic threshold range and the secondary dynamic threshold range of each monitoring parameter are associated and integrated to construct the fault judgment threshold system.

3. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 1, characterized in that, The step of generating the time sequence monitoring data set comprises: Parameter configuration information matching the current motion mode is received, a real-time data acquisition process is started, and monitoring parameter data and dynamic physical parameters of the core cooperative link in a loading cycle process of the dynamic compression and shear testing machine are acquired; Based on a loading cycle control signal of the dynamic compression and shear testing machine, a start node and a termination node of each loading cycle are identified, the monitoring parameter data and the dynamic physical parameters are segmented according to the loading cycle turns to form segmented data groups; According to the time sequence order of the loading cycles, a loading cycle serial number identifier is assigned to each segmented data group; the loading cycle serial number identifier comprises a loading cycle sequence code and a loading cycle start and end timestamp; The monitoring parameter segmented data and the dynamic physical parameter segmented data under the same loading cycle serial number identifier are bound, and validity verification is performed to generate the time sequence monitoring data set.

4. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 2, characterized in that, The specific steps of constructing the fault judgment reference library comprise: A reference candidate range is configured, loading cycle segmented data groups in the reference candidate range in the time sequence monitoring data set are extracted, and are integrated to form a reference candidate data set; The fault judgment threshold system is called, the primary dynamic threshold range corresponding to each monitoring parameter is extracted as a running state judgment basis, a candidate cycle full-amount normality detection process is executed to judge whether the loading cycle is a normal loading cycle; If all the loading cycles in the reference candidate range are judged to be normal loading cycles, the reference candidate data set is confirmed to be valid, and a fault judgment reference establishment operation is triggered: Based on the benchmark candidate dataset, the characteristic values of each core collaborative link are extracted respectively, including the monitoring parameter characteristic values and the dynamic physical parameter characteristic values, and the inherent physical parameters of the core collaborative link are called as the benchmark static attributes for supplement; The characteristic dimensions are divided according to the core collaborative links, the monitoring parameter characteristic values, the dynamic physical parameter characteristic values and the inherent physical parameters under the same link are associated and bound, and the benchmark characteristic subsets corresponding to each core collaborative link are generated; The benchmark characteristic subsets of all core collaborative links are integrated to form the fault judgment benchmark library corresponding to the current motion mode.

5. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 2, characterized in that, The steps of generating the prediction values and the confidence interval of each monitoring parameter in the subsequent loading cycle according to the deviation trend prediction include: The segmented data groups of the subsequent loading cycle in the time sequence monitoring data set are continuously received, the first-level dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system is called, and it is determined whether the current loading cycle is a normal loading cycle, if yes, the deviation analysis operation is triggered; The monitoring parameters and the dynamic physical parameters of the normal loading cycle are extracted, and the historical normal loading cycle segmented data groups of the preset monitoring period before the current loading cycle are traced back, the corresponding historical monitoring parameters and historical dynamic physical parameters are extracted; The segmented data groups of the current loading cycle and the historical normal loading cycle are subjected to smoothing filtering pretreatment and data alignment operation, and the data set for prediction analysis is integrated; The benchmark characteristic subsets corresponding to the core collaborative links are matched according to the core collaborative link dimensions by calling the fault judgment benchmark library, the deviation data calculation is performed based on the data set for prediction analysis, the deviation data includes the instantaneous deviation amount, the cycle cumulative deviation amount and the deviation change rate, and the deviation analysis matrix is formed; The deviation data of the current loading cycle and the historical normal loading cycle are extracted from the deviation analysis matrix, and the historical deviation time sequence data set is arranged in the order of time sequence; The deviation trend prediction model is constructed in advance, the historical deviation time sequence data set is input into the deviation trend prediction model, and the deviation prediction value and the confidence interval of the monitoring parameter in the subsequent loading cycle are output.

6. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 5, characterized in that, The deviation trend prediction model is composed of a data input layer, a bidirectional LSTM feature extraction layer, an attention weight distribution layer, a full connection prediction layer and an output layer; The data input layer is used for receiving the historical deviation time sequence data set; The bidirectional LSTM feature extraction layer is used for mining the bidirectional time sequence correlation features of the deviation data of the historical deviation time sequence data set, and obtaining a deviation feature matrix; The attention weight distribution layer is used for differentiating the weight distribution of the deviation feature matrix output by the bidirectional LSTM, and obtaining a weighted feature vector; The full connection prediction layer is used for nonlinear transformation and dimension mapping of the weighted feature vector output by the attention weight distribution layer, and mapping the weighted feature vector output by the attention layer into the deviation prediction value of the subsequent loading cycle; The output layer is used for integrating the prediction results of the full connection layer, and outputting the deviation prediction value and the confidence interval thereof.

7. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 5, characterized in that, The steps of generating the compensation instruction according to the deviation trend type and the deviation level include: Based on the values of the monitoring parameters in the current loading cycle and the predicted values and confidence intervals of the monitoring parameters in the subsequent loading cycle, the predicted values and confidence intervals of the monitoring parameters in the subsequent loading cycle are generated; The dynamic threshold range corresponding to each monitoring parameter in the fault judgment threshold system is called to determine the deviation state of each monitoring parameter. The deviation state includes no compensation requirement, ordinary compensable deviation, critical compensable deviation, and non-compensable fault deviation. For the monitoring parameters determined as ordinary compensable deviation and critical compensable deviation, multi-dimensional attribute information is extracted based on the deviation analysis matrix, including link attribution attribute, deviation magnitude attribute, and trend evolution attribute, and is integrated into a standardized deviation attribute label. A preset compensation strategy library is called, which has a three-dimensional mapping rule of link attribution attribute, deviation magnitude attribute, and trend evolution attribute, and is used to match a compensation strategy according to the multi-dimensional attribute information. The matched compensation strategy is converted into an executable standardized compensation instruction, which is sent to the corresponding execution mechanism to perform the compensation operation. After the subsequent loading cycle is completed, the compensated monitoring parameters are extracted. If the compensated monitoring parameters are within the secondary dynamic threshold range, it is determined that the compensation is effective. If the compensated monitoring parameter deviation is outside the secondary dynamic threshold range, a targeted compensation strategy is regenerated and executed. The number of times that the targeted compensation strategy is regenerated and executed is counted. If it is greater than a preset control threshold, it is determined that the compensation is ineffective, and a compensation abnormality warning is given.

8. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 2, characterized in that, The step of identifying the potential fault monitoring parameter and constructing a potential fault monitoring parameter list comprises: Receiving the actual value dataset of each monitoring parameter in the current loading cycle and the predicted value and confidence interval of each monitoring parameter in the subsequent loading cycle, and calling the fault judgment threshold system corresponding to the current motion mode to associate the link attribution identifier and the fault sensitive point position information of each parameter; Binding the monitoring parameters according to the structure of monitoring parameter ID, link attribution identifier, fault sensitive point position, value type, and dynamic threshold range to generate a judgment dataset; Based on the link attribution identifier of the judgment dataset, the monitoring parameters are split into multiple core collaborative sub-link parameter groups according to the core collaborative link. The monitoring parameters in each core collaborative sub-link parameter group are further associated with the monitoring parameter attributes of the corresponding fault sensitive points; For the monitoring parameters in each core collaborative sub-link parameter group after splitting, the coupling determination of actual value overrun state and predicted value overrun trend is performed for each monitoring parameter, including: If the actual value of a monitoring parameter in the current loading cycle exceeds the corresponding first dynamic threshold range, it is determined as a potential fault monitoring parameter, and the overrun type is marked as actual value overrun; If the actual value of a monitoring parameter in the current loading cycle does not exceed the first dynamic threshold range, but the predicted value in the subsequent loading cycle exceeds the first dynamic threshold range, or the confidence interval intersects with the first dynamic threshold range, the trend tracing determination is started. According to the time sequence change slope of the historical actual value data of the monitoring parameter in the historical loading cycle, it is determined whether the monitoring parameter is a potential fault monitoring parameter. If yes, the overrun type is marked as predicted value overrun. The monitoring parameter ID, the link belonging identifier, the fault sensitive point position, the overrun type, and the overrun amplitude of the potential fault monitoring parameter are used to construct a potential fault monitoring parameter list.

9. The dynamic compression-shear testing machine operating state online monitoring and fault early warning system according to claim 1, characterized in that, The step of calculating the final fault probability of each core collaborative sublink comprises: Taking the physical characteristics of each potential fault monitoring parameter as input, the triggering condition of the corresponding fault sensitive point node in the link-based fault causal diagram is matched, and the node triggering probability is calculated; Based on the preset propagation probability of the directed edge between the fault sensitive point node and the core collaborative sublink node, the single path comprehensive fault probability of the corresponding core collaborative sublink of a single fault sensitive point node is obtained by recursive calculation along the fault propagation direction, as the propagation probability contribution value of the potential fault monitoring parameter with respect to the corresponding core collaborative sublink; The propagation probability contribution value refers to the triggering probability of the fault sensitive point node corresponding to a potential fault monitoring parameter in the same core collaborative sublink, which is weighted by the fault propagation probability between nodes, and is a quantitative contribution degree to the final fault probability of the core collaborative sublink; The node triggering probability and the propagation probability contribution value corresponding to all potential fault monitoring parameters in the same core collaborative sublink are integrated to obtain the final fault probability of each core collaborative sublink.

Images