Self-adaptive anomaly identification method and system for operation condition of heading machine

Abstract

The invention provides a self-adaptive anomaly identification method and system for the operation condition of a heading machine, and relates to the technical field of mining engineering tunneling, and the method comprises the steps: driving a plurality of associated devices to execute working condition collection based on a preset cooperative sampling rule when a lithology sudden change event is captured, carrying out the time sequence feature matching backtracking of the lithology sudden change event, and obtaining a lithology sudden change event; and dynamically calling an equipment association judgment model according to a real-time lithologic transition stage, calculating a correlation coefficient matrix, loading the correlation coefficient matrix to the equipment association judgment model to carry out space-time association coupling judgment, and when a judgment result is association coupling failure, analyzing a failure feature vector to carry out fault pointing probability matching, and outputting a real-time fault equipment group. The technical problem of insufficient timeliness and accuracy of abnormal recognition caused by monitoring the state of the heading machine through single or fixed parameters in the prior art is solved. The technical effects of improving the comprehensiveness, real-time performance and accuracy of abnormal recognition of the heading machine and guaranteeing efficient and safe operation of the heading machine are achieved.

Description

Technical Field

[0001] This invention relates to the field of tunneling technology in mining engineering, and more specifically to an adaptive anomaly identification method and system for tunneling machine operating conditions. Background Technology

[0002] In the field of tunneling operations in mining engineering, tunneling machines (TBMs) are core construction equipment, and the stability and reliability of their operating conditions directly affect the progress and quality of the project. However, traditional TBM anomaly identification technologies are often limited to the monitoring and analysis of single equipment or fixed parameters, failing to fully consider the dynamic impact of lithological mutations—a key variable—on the operating status of equipment under complex geological conditions. Lithological mutations can cause rapid changes in the operating conditions of equipment, and traditional technologies struggle to capture these changes in real time and accurately identify the relationships between equipment, resulting in insufficient timeliness and accuracy in anomaly identification, making it difficult to meet the demands of efficient and safe tunneling operations.

[0003] Existing technologies suffer from problems such as insufficient timeliness and accuracy in anomaly identification due to the use of single or fixed parameters to monitor the status of tunneling machines. Summary of the Invention

[0004] This application provides an adaptive anomaly identification method and system for tunneling machine operating conditions, which addresses the technical problem that existing technologies use single or fixed parameters to monitor the tunneling machine status, resulting in insufficient timeliness and accuracy of anomaly identification.

[0005] In view of the above problems, this application provides an adaptive anomaly identification method and system for tunneling machine operating conditions.

[0006] The first aspect of this application provides an adaptive anomaly identification method for the operating conditions of a tunneling machine. The method includes: when a lithological abrupt change event is captured, triggering the acquisition of operating conditions for multiple pre-set associated devices within the tunneling machine; based on pre-set collaborative sampling rules, driving multiple distributed edge acquisition nodes of the multiple associated devices to perform millisecond-level synchronous operating condition acquisition, generating multi-channel operating condition time-series data; performing time-series feature matching and backtracking on the lithological abrupt change event to locate the real-time lithological transition stage; dynamically retrieving an equipment association judgment model based on the real-time lithological transition stage; calculating a cross-correlation matrix based on the multi-channel operating condition time-series data, and loading the cross-correlation matrix into the equipment association judgment model for spatiotemporal association coupling determination; if the spatiotemporal association coupling determination result is an association coupling failure, then parsing a failure feature vector from the cross-correlation matrix; performing fault pointing probability matching based on the failure feature vector, and outputting a real-time faulty equipment group, wherein the real-time faulty equipment group carries an adaptive graded fault level identifier.

[0007] Preferably, the real-time lithological transition stage is located by performing time-series feature matching and backtracking on the lithological abrupt change event. The method includes: using the timestamp of the lithological abrupt change event as a reference point, calling the time-series data before and after a preset backtracking time window to obtain a multi-channel equipment operating condition time-series segment; outputting multi-modal lithological response features by solving the multi-channel equipment operating condition time-series segment, wherein the multi-modal lithological response features include the propulsion speed reduction rate, the cutterhead torque fluctuation amplitude, and the proportion of vibration signal main frequency energy; and locating the real-time lithological transition stage by using the multi-modal lithological response features to traverse and compare multiple standard modal feature templates of multiple standard transition stages in the lithological transition feature library.

[0008] Preferably, the method for dynamically retrieving equipment association judgment models based on the real-time lithological transition stage includes: interactively obtaining multiple equipment parameter association rules for the multiple standard transition stages; configuring multiple multi-task comparison engines for the multiple equipment parameter association rules, configuring multiple distributed parallel processing architectures for the multiple multi-task comparison engines, and completing the construction of multiple standard association judgment models; using the multiple standard transition stages as indexes to index and store the multiple standard association judgment models, generating an association judgment model library; and dynamically retrieving the equipment association judgment model from the association judgment model library based on the real-time lithological transition stage.

[0009] Preferably, after calculating the cross-correlation matrix based on the multi-channel operating condition time-series data, the cross-correlation matrix is ​​loaded into the equipment association judgment model for spatiotemporal association coupling determination. The method includes: combining and enumerating the multi-channel operating condition time-series data to obtain multiple sets of dual-channel operating condition time-series data; using a preset sliding segmentation window and sliding segmentation step size as the calculation basis, calculating the maximum cross-correlation number and operating condition delay of the multiple sets of dual-channel operating condition time-series data, and outputting multiple symmetric relation matrices; aggregating the multiple symmetric relation matrices according to the channel pair topology of the multiple sets of dual-channel operating condition time-series data to generate the cross-correlation matrix; loading the cross-correlation matrix into the equipment association judgment model for spatiotemporal association coupling determination, and outputting the spatiotemporal association coupling determination result.

[0010] Preferably, the method for parsing the failure feature vector from the cross-correlation matrix includes: extracting multiple channel pair relationship data fragments from the cross-correlation matrix according to the device parameter association rules of the device association judgment model; loading the multiple channel pair relationship data fragments into the multi-task comparison engine of the device association judgment model, performing parallel spatiotemporal coupled dual-threshold decision-making, and outputting multiple failure state sets; dynamically aggregating the multiple failure state sets according to the rule ID index of the multiple channel pair relationship data fragments, and outputting the failure feature vector.

[0011] Preferably, when a lithological mutation event is captured, the operating condition data acquisition of multiple pre-set associated devices in the tunneling machine is triggered. The method includes: demodulating the tunneling machine's propulsion speed time-series signal using a sliding window based on a sampling window, and outputting the propulsion speed standard deviation; performing local extreme value density analysis on the tunneling machine's cutterhead torque signal, and outputting the cutterhead torque extreme value density, wherein the propulsion speed time-series signal demodulation and the cutterhead torque signal analysis are performed synchronously; when the propulsion speed standard deviation suddenly increases by more than 30% of the historical mean baseline, satisfying P sampling windows, and the cutterhead torque extreme value density reaches more than Q times the historical mean density, a lithological mutation event triggering command is generated; the lithological mutation event triggering command is sent to the multiple distributed edge acquisition nodes based on a unified timestamp, triggering the acquisition of the operating condition data of the multiple associated devices.

[0012] Preferably, the method for performing fault-pointing probability matching based on the failure feature vectors to output a real-time fault device group includes: interactively obtaining multiple sets of sample feature vectors of the multiple associated devices under multiple sets of sample fault states; constructing a three-level index tree based on the multiple associated devices, multiple sets of sample fault states, and multiple sets of sample feature vectors; loading the failure feature vectors into the three-level index tree, performing coarse screening of fault states based on pre-selected similarity to obtain a candidate fault state set; performing fine matching of the failure feature vectors and the candidate fault state set based on Euclidean distance to output multiple device fault association probability pairs; and performing device fault descending classification based on the multiple device fault association probability pairs to output the real-time fault device group, wherein the real-time fault device group has an adaptive classification fault level identifier.

[0013] Preferably, the adaptive anomaly identification method for tunneling machine operating conditions further includes: dynamically updating the historical mean baseline and historical mean density based on a preset cumulative tunneling mileage.

[0014] Preferably, the adaptive anomaly identification method for tunneling machine operating conditions further includes: constructing a fault verification priority instruction for the real-time fault equipment group based on the adaptive graded fault level identifier, and sending the fault verification priority instruction to the ground central control system.

[0015] A second aspect of this application provides an adaptive anomaly identification system for the operating conditions of a tunneling machine. The system includes: a condition acquisition triggering module, used to trigger the acquisition of operating conditions of multiple pre-set associated devices in the tunneling machine when a lithological abrupt change event is captured; a collaborative sampling execution module, used to drive multiple distributed edge acquisition nodes of the multiple associated devices to perform millisecond-level synchronous condition acquisition based on pre-set collaborative sampling rules, generating multi-channel operating condition time-series data; a time-series feature backtracking module, used to perform time-series feature matching backtracking on the lithological abrupt change event to locate the real-time lithological transition stage; and a model retrieval module, used to retrieve the model based on the real-time... During the lithological transition stage, the equipment association judgment model is dynamically retrieved; the association coupling decision module is used to calculate the cross-correlation matrix based on the multi-channel operating condition time series data, and then load the cross-correlation matrix into the equipment association judgment model to perform spatiotemporal association coupling judgment; the failure feature analysis module is used to determine whether the spatiotemporal association coupling judgment result is an association coupling failure, and then parse the failure feature vector from the cross-correlation matrix; the fault output module is used to perform fault pointing probability matching based on the failure feature vector and output a real-time fault equipment group, wherein the real-time fault equipment group has an adaptive graded fault level identifier.

[0016] One or more technical solutions provided in this application have at least the following technical effects or advantages: The method provided in this application, when capturing a lithological mutation event, triggers the acquisition of operating conditions of multiple pre-set associated devices in the tunneling machine; based on pre-set collaborative sampling rules, it drives multiple distributed edge acquisition nodes of the multiple associated devices to perform millisecond-level synchronous operating condition acquisition, generating multi-channel operating condition time-series data; it performs time-series feature matching and backtracking on the lithological mutation event to locate the real-time lithological transition stage; based on the real-time lithological transition stage, it dynamically retrieves the device association judgment model; after calculating the cross-relation matrix based on the multi-channel operating condition time-series data, it loads the cross-relation matrix into the device association judgment model for spatiotemporal association coupling determination; if the spatiotemporal association coupling determination result is association coupling failure, it parses the failure feature vector from the cross-relation matrix; based on the failure feature vector, it performs fault pointing probability matching and outputs the real-time faulty device group. This achieves the technical effect of improving the comprehensiveness, real-time performance, and accuracy of tunneling machine anomaly identification, ensuring the efficient and safe operation of the tunneling machine. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the adaptive anomaly identification method for tunneling machine operating conditions provided in this application.

[0018] Figure 2 This is a schematic diagram of the adaptive anomaly identification system for tunneling machine operating conditions provided in this application.

[0019] Figure labeling: 10 for working condition acquisition triggering module, 20 for collaborative sampling execution module, 30 for time series feature backtracking module, 40 for model retrieval module, 50 for correlation coupling decision module, 60 for failure feature analysis module, and 70 for fault output module. Detailed Implementation

[0020] This application provides an adaptive anomaly identification method and system for tunneling machine operating conditions, addressing the technical problem that existing technologies using single or fixed parameters to monitor tunneling machine status result in insufficient timeliness and accuracy in anomaly identification. It achieves the technical effect of improving the comprehensiveness, real-time performance, and accuracy of tunneling machine anomaly identification, ensuring the efficient and safe operation of the tunneling machine.

[0021] The technical solutions of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. It should be understood that the present invention is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention. It should also be noted that, for ease of description, only the parts related to the present invention are shown in the accompanying drawings, not all of them.

[0022] Example 1, as Figure 1 As shown, this application provides an adaptive anomaly identification method for tunneling machine operating conditions, the method comprising: When a lithological mutation event is captured, the operating conditions of multiple pre-set associated devices in the tunneling machine are collected.

[0023] Specifically, the operating conditions of tunneling machines (TBMs) are closely related to lithological abrupt changes. When a TBM cuts different rock strata, its operating parameters such as propulsion speed, cutterhead torque, and vibration are directly affected by the hardness, structure, and integrity of the rock strata. When lithology changes abruptly, the ground resistance and cutting conditions change instantaneously, causing significant fluctuations in these operating parameters. Therefore, by capturing lithological abrupt changes, the operating conditions of multiple pre-set associated devices in the TBM are collected. These pre-set associated devices are those that are related to the operating conditions of the TBM's main drive system, cutting system, propulsion system, and support system. These associated devices include, but are not limited to, the cutting motor, propulsion cylinder, and conveyor. Based on this, rapid identification and response to geological changes can be achieved, thereby ensuring accurate, comprehensive, and timely collection of the operating conditions of the TBM equipment.

[0024] Furthermore, when a lithological mutation event is captured, the operating conditions of multiple pre-set associated devices in the tunneling machine are collected. The method includes: demodulating the tunneling machine's propulsion speed time-series signal using a sliding window based on a sampling window, and outputting the propulsion speed standard deviation; performing local extreme value density analysis on the tunneling machine's cutterhead torque signal, and outputting the cutterhead torque extreme value density, wherein the propulsion speed time-series signal demodulation and the cutterhead torque signal analysis are performed simultaneously; when the propulsion speed standard deviation suddenly increases by more than 30% of the historical mean baseline, satisfying P sampling windows, and the cutterhead torque extreme value density reaches more than Q times the historical mean density, a lithological mutation event triggering command is generated; the lithological mutation event triggering command is sent to the multiple distributed edge acquisition nodes based on a unified timestamp, triggering the collection of operating conditions of the multiple associated devices.

[0025] Specifically, a sampling window refers to a time interval during the continuous acquisition of tunneling machine operating condition signals, where the signal is segmented into segments with a fixed time length or a fixed number of data points. This segment is used to calculate local signal characteristic values, such as the standard deviation of the advance speed or the extreme value density of the cutterhead torque. The length of the sampling window is usually set according to the typical time scale of the tunneling machine's operating state changes. For example, for operating conditions with faster tunneling speeds and more frequent changes in geological conditions, the sampling window can be appropriately smaller, such as 0.5-2 seconds, to more sensitively capture signal changes. For operating conditions with slower tunneling speeds and relatively stable geological conditions, the sampling window can be larger, such as 2-5 seconds, to reduce noise interference and improve calculation stability. Based on a preset window, sliding window standard deviation demodulation is performed on the time-series signal of the tunneling machine's propulsion speed. The propulsion speed time-series signal refers to a continuous data sequence formed by the change of the distance moved by the propulsion device per unit time during the tunneling operation. The propulsion speed time-series signal can reflect the propulsion speed of the tunneling machine at different times in real time. Sliding window standard deviation demodulation refers to gradually sliding a fixed-length time window on the signal sequence and calculating the standard deviation of the data in each window. The trend of the standard deviation reflects the fluctuation range of the propulsion speed, thereby quantitatively analyzing the sensitivity to formation disturbance and obtaining the standard deviation of the propulsion speed. The larger the standard deviation of the propulsion speed, the more severe the fluctuation of the propulsion speed. Simultaneously, local extremum density analysis is performed on the tunneling machine cutterhead torque signal. The tunneling machine cutterhead torque signal refers to the data set consisting of the torque acting on the cutterhead over time as it rotates and cuts rock strata. The cutterhead torque signal reflects the magnitude of the resistance experienced by the cutterhead during cutting. Local extremum density refers to the ratio of the number of occurrences of local maximum or minimum values ​​of the torque signal to the duration within a given time window. It reflects the frequency of load fluctuations experienced by the cutterhead under changes in rock strata. Through local extremum density analysis of the cutterhead torque signal, the key indicator of cutterhead torque extremum density is output. A higher cutterhead torque extremum density value indicates more frequent occurrences of local extrema in the cutterhead torque, suggesting more frequent and drastic fluctuations in the cutterhead torque within a short period, reflecting the extreme instability of the load experienced by the cutterhead during rock cutting. The demodulation of the standard deviation of the propulsion speed and the cutterhead torque extremum density analysis are performed simultaneously. This parallel processing ensures high consistency between the two key indicators over time, enabling parallel monitoring of propulsion stability and cutting load fluctuations.

[0026] When the standard deviation of the advance speed increases by more than 30% relative to the historical mean baseline within P consecutive sampling windows, and the extreme value density of the cutterhead torque reaches more than Q times the historical mean density, the current operating state is determined to meet the triggering conditions for a lithological mutation event. A lithological mutation event triggering command is then generated immediately. The historical mean baseline is a long-term average value dynamically calculated based on a preset cumulative tunneling mileage, used to adapt to benchmark changes at different construction stages. The historical mean density corresponds to the frequency benchmark of local extreme values ​​under long-term stable operating conditions. P and Q are both positive integers, with P greater than or equal to 3 and Q greater than P. After the triggering command is generated, it is sent with a unified timestamp to multiple distributed edge acquisition nodes located at various positions on the tunneling machine. This allows each acquisition node to simultaneously initiate the acquisition of operating conditions for multiple related devices, including but not limited to the cutting motor, propulsion cylinder, and conveyor. By determining whether a lithological mutation event triggers the acquisition of operating conditions for multiple preset related devices in the tunneling machine, comprehensive, timely, and accurate operating information of key equipment can be obtained.

[0027] Based on preset collaborative sampling rules, multiple distributed edge acquisition nodes of the multiple associated devices are driven to perform millisecond-level synchronous operating condition acquisition and generate multi-channel operating condition time series data.

[0028] Specifically, upon receiving a lithological mutation event trigger command, multiple distributed edge acquisition nodes on various associated devices are driven to initiate operational condition data acquisition based on preset collaborative sampling rules. These preset collaborative sampling rules are sampling strategies constructed by comprehensively considering the physical connections, signal transmission characteristics, and business logic relationships between the various associated devices of the tunneling machine. This ensures the temporal and logical consistency of multi-source data. For example, the tunneling machine's cutting motor, propulsion cylinder, and conveyor interact and work collaboratively during operation. Changes in propulsion speed directly affect the feed rate of the cutting head, while the stress on the cutting head is fed back to the load on the propulsion cylinder. Simultaneously, the conveyor's operating speed needs to match the cutting motor's discharge speed. Furthermore, based on these relationships, the preset collaborative sampling rules clearly define the operational parameters that each associated device needs to collect under different operating conditions, as well as key information such as the collection frequency and time interval, comprehensively reflecting the equipment's operating status and performance. Distributed edge acquisition nodes are hardware units installed near various associated devices, possessing local signal acquisition and preliminary processing capabilities. Through a low-latency industrial bus or time synchronization network, they achieve high-precision clock alignment with the main control unit, enabling millisecond-level synchronous start-up of acquisition upon receiving a trigger command with a unified timestamp. Millisecond-level synchronization emphasizes that the deviation of the data acquisition start points of different devices on the time axis does not exceed milliseconds. In this way, each distributed edge acquisition node simultaneously acquires various types of operating condition signals, such as propulsion speed, cutterhead torque, hydraulic pressure, vibration acceleration, and motor current, and arranges them in chronological order to form multi-channel operating condition time-series data containing multiple data channels. Multi-channel refers to parallel data streams from different physical quantities and different devices, comprehensively reflecting the operating status of the tunneling machine within a certain time period.

[0029] The lithological abrupt change events are backtracked by time-series feature matching to locate the real-time lithological transition stage.

[0030] Specifically, for detected lithological abrupt events, feature data such as propulsion speed and cutterhead torque are extracted based on multi-channel operating condition time series data. The feature data are then matched with typical features of different lithological transition stages in historical data. The stage with the highest similarity is determined based on the degree of matching, thereby accurately locating the current real-time lithological transition stage.

[0031] Furthermore, the method involves performing time-series feature matching and backtracking on the lithological abrupt change event to locate the real-time lithological transition stage. This includes: using the timestamp of the lithological abrupt change event as a reference point, retrieving time-series data before and after a preset backtracking time window to obtain multi-channel equipment operating condition time-series segments; outputting multi-modal lithological response features by solving the multi-channel equipment operating condition time-series segments, wherein the multi-modal lithological response features include the propulsion speed reduction rate, cutterhead torque fluctuation amplitude, and the proportion of vibration signal dominant frequency energy; and using the multi-modal lithological response features to traverse and compare multiple standard modal feature templates of multiple standard transition stages in the lithological transition feature library to locate the real-time lithological transition stage.

[0032] Specifically, after a lithological abrupt change event is triggered and multi-channel operating condition data acquisition is completed, the detected lithological abrupt change event is backtracked using time-series feature matching to locate the current real-time lithological transition stage of the tunnel boring machine (TBM). First, using the unified timestamp of the lithological abrupt change event as a reference point, multi-channel equipment operating condition data segments covering a certain time period before and after that timestamp are retrieved. The preset backtracking time window is a time range set according to the typical lag characteristics of the TBM's response to strata changes, including both the signal change trend before the event and the dynamic response after the event, thus ensuring the completeness of feature extraction. Subsequently, the extracted multi-channel data segments are processed using a specific algorithm model, such as a time-series analysis algorithm based on signal processing, to output multimodal lithological response features. Multimodal lithological response features refer to a set of features that integrate multiple signal types from different physical quantities and different equipment, reflecting the changes in rock strata characteristics during the tunneling process. Multimodal lithological response characteristics include: the rate of decrease in advance speed, the amplitude of cutterhead torque fluctuation, and the proportion of dominant frequency energy in the vibration signal. The rate of decrease in advance speed, i.e., the relative decrease in advance speed per unit time, reflects the degree of increase in formation cutting resistance. The amplitude of cutterhead torque fluctuation, i.e., the peak-to-valley difference of the torque signal within the analysis window, is sensitive to changes in lithological hardness and heterogeneity. The proportion of dominant frequency energy in the vibration signal, i.e., the proportion of energy in the dominant frequency band of the vibration signal to the total energy, reveals changes in impact characteristics during the cutting process. Finally, the above multimodal lithological response characteristics are systematically compared with standard modal feature templates of multiple pre-established standard transition stages in the lithological transition feature library. The lithological transition feature library is a set of phased features constructed based on historical construction data and typical formation test data, covering different types of lithological transition modes, such as from soft rock to hard rock and from hard rock to fractured zones. Through similarity calculations, such as Euclidean distance and cosine similarity methods, the standard transition stage corresponding to the current multimodal characteristics is accurately matched, thereby achieving precise positioning of the real-time lithological transition stage.

[0033] Based on the real-time lithological transition stage, the equipment association judgment model is dynamically retrieved.

[0034] Specifically, based on the identified real-time lithological transition stage, a suitable equipment association judgment model is dynamically retrieved from a pre-set model library. This model comprehensively considers various key parameters and interrelationships of equipment operation during this lithological transition stage, enabling accurate and reliable assessment of the equipment's current state. This provides a reliable basis for subsequent adjustments to the tunneling machine's operation, ensuring efficient and safe operation.

[0035] Furthermore, based on the real-time lithological transition stage, the device association judgment model is dynamically retrieved. The method includes: interactively obtaining multiple device parameter association rules for the multiple standard transition stages; configuring multiple multi-task comparison engines for the multiple device parameter association rules, configuring multiple distributed parallel processing architectures for the multiple multi-task comparison engines, and completing the construction of multiple standard association judgment models; using the multiple standard transition stages as indexes, indexing and storing the multiple standard association judgment models to generate an association judgment model library; and dynamically retrieving the device association judgment model from the association judgment model library based on the real-time lithological transition stage.

[0036] Specifically, by leveraging expert experience, historical construction data, and experimental verification results, multiple equipment parameter association rules corresponding to standard lithological transition stages are interactively obtained. A standard transition stage refers to a predefined typical stage of transition between different lithologies, such as from soft rock to medium-hard rock, or from medium-hard rock to hard rock. Equipment parameter association rules refer to the correlation patterns of operating parameters between different devices under specific lithological transition stages, including parameter coupling direction, correlation coefficient range, and response delay characteristics. A multi-task comparison engine is configured for multiple equipment parameter association rules, implemented by mapping each association rule to an independent comparison task within the engine. Each task includes a specific combination of input data channels, feature calculation methods, and threshold judgment conditions. For example, if a rule requires monitoring "the correlation coefficient between the cutterhead torque and the propulsion cylinder pressure to be no less than 0.85 and the time delay no more than 200ms" during a certain lithological transition stage, a task can be created in the engine, specifying the input channels as the cutterhead torque signal and the propulsion cylinder pressure signal, calculating the maximum cross-correlation coefficient and phase delay of the sliding window, and setting corresponding thresholds for real-time judgment. For another rule, such as "the fluctuation amplitude of the main drive motor current and the change in the proportion of the cutterhead vibration main frequency energy must be positively correlated," another task can be configured, using the current fluctuation amplitude and the proportion of vibration energy as input, and using a trend correlation analysis algorithm for judgment. In this way, the multi-task comparison engine can run the calculation tasks corresponding to multiple rules in parallel at the same time, realizing the synchronous analysis and judgment of the relationship between multiple equipment parameters. The multi-task comparison engine refers to a computing module capable of processing multiple tasks simultaneously. Employing advanced multi-threading or distributed computing technologies, it can efficiently compare and analyze equipment parameters under different rules. It can be implemented using existing multivariate parallel analysis frameworks in industrial automation or condition monitoring systems, such as multi-channel correlation analysis engines used in SCADA or DCS systems, multi-channel vibration feature comparison algorithms used in condition monitoring systems, or multi-feature matching pipelines deployed using parallel computing frameworks like MATLAB Parallel Computing or Apache Spark. After configuration, multiple comparison engines are deployed in a distributed parallel processing architecture, allowing each engine to run simultaneously on multiple computing nodes, significantly improving processing speed and efficiency. This enables rapid processing of large amounts of equipment parameter data, thereby completing the construction of multiple standard correlation judgment models. Based on the above configuration, a standard correlation judgment model is constructed for each standard lithological transition stage, and the standard lithological transition stages are used as an index for indexed storage, forming a correlation judgment model library. The index is similar to a book's table of contents, allowing for quick location of specific standard correlation judgment models and improving model retrieval efficiency.Finally, once the real-time lithological transition stage is identified and matched, the corresponding equipment association judgment model is dynamically retrieved directly from the association judgment model library through the index. This avoids the accuracy loss of the general model under specific stages, thereby timely and accurately assessing the operating status of the equipment under the current lithological conditions. This enables the diagnostic process to adapt to formation changes and ensures that the optimal analysis model can be called for judgment under different lithological transition conditions.

[0037] After calculating the cross-correlation matrix based on the multi-channel operating condition time-series data, the cross-correlation matrix is ​​loaded into the equipment association judgment model for spatiotemporal association coupling judgment.

[0038] Specifically, based on the multi-channel operating condition time series data, the sliding window calculation is first performed on each pair of channels to obtain the maximum number of cross-correlation numbers and corresponding time delays for each pair of channels in each window. The results are then aggregated into a cross-correlation number matrix according to the channel topology order. This cross-correlation number matrix is ​​used as a feature input into the device association judgment model, and the spatiotemporal association coupling judgment result is output.

[0039] Furthermore, after calculating the cross-correlation matrix based on the multi-channel operating condition time-series data, the cross-correlation matrix is ​​loaded into the equipment association judgment model for spatiotemporal association coupling determination. The method includes: combining and enumerating the multi-channel operating condition time-series data to obtain multiple sets of dual-channel operating condition time-series data; using a preset sliding segmentation window and sliding segmentation step size as the calculation basis, calculating the maximum cross-correlation number and operating condition delay of the multiple sets of dual-channel operating condition time-series data, and outputting multiple symmetric relation matrices; aggregating the multiple symmetric relation matrices according to the channel pair topology of the multiple sets of dual-channel operating condition time-series data to generate the cross-correlation matrix; loading the cross-correlation matrix into the equipment association judgment model for spatiotemporal association coupling determination, and outputting the spatiotemporal association coupling determination result.

[0040] Specifically, the multi-channel operating condition time series data is combined and enumerated, that is, all possible channel pairs are exhaustively enumerated in a pairwise manner to obtain multiple sets of dual-channel time series data for relation measurement. Each set of dual-channel data is segmented and analyzed based on a preset sliding segmentation window and sliding segmentation step size. The sliding segmentation window is an analysis interval that slides continuously at a fixed length from the original time series, and the sliding segmentation step size is the displacement length of adjacent analysis intervals, determining the distance the window slides each time. Based on a preset sliding window and sliding step size, the maximum cross-correlation coefficient and corresponding operating delay of multiple sets of dual-channel operating condition time series data are calculated within each window. The maximum cross-correlation coefficient is calculated, for example, by multiplying two dual-channel time series data sequences by different time offsets and then summing the results, obtaining a series of cross-correlation values. The maximum value, after normalization, is the maximum cross-correlation coefficient, reflecting the similarity between the two channel data at different time offsets. The value ranges from -1 to 1; the closer the absolute value is to 1, the stronger the correlation. During the calculation of the maximum cross-correlation coefficient, the time offset of the two time series data sequences when the maximum cross-correlation value is obtained is recorded. This offset is the operating delay, reflecting the time delay of one channel data change relative to the other. The results from each window are aggregated to form a symmetric relation matrix for each channel pair. Each symmetric relation matrix corresponds to a set of dual-channel data, and the elements of the symmetric relation matrix reflect the maximum cross-correlation coefficient and operating delay information of this pair of channel data.

[0041] Channel topology refers to the structural relationships, such as the physical location or functional associations of different channels within a tunneling machine, which can be obtained from equipment design documents. Based on these channel topology relationships, multiple symmetrical relationship matrices are aggregated, integrating scattered channel relationship information into a single matrix that comprehensively reflects the overall association characteristics of the equipment. For example, aggregation can be performed using K-means clustering or analytic hierarchy process (AHP) based on channel correspondence and functional affinity to obtain a cross-correlation matrix. This cross-correlation matrix integrates the maximum cross-correlation coefficients and operational delay information across all channels, comprehensively presenting the association characteristics between each channel. This cross-correlation matrix is ​​then loaded into the equipment association judgment model. This model, combined with pre-defined decision rules and algorithms, analyzes the association coupling of the equipment in different time and spatial dimensions, determining whether the associations between different parts of the equipment are normal and whether there are potential fault risks. Finally, it outputs a spatiotemporal association coupling judgment result, enabling adaptive anomaly identification and fault prediction during tunneling machine operation, improving real-time diagnostic capabilities and construction safety under complex geological conditions.

[0042] If the spatiotemporal correlation coupling determination result is a correlation coupling failure, then the failure feature vector is parsed from the cross-correlation matrix.

[0043] Specifically, when the spatiotemporal correlation coupling judgment result indicates a failure in correlation coupling, it means that during the operation of the tunneling machine, the correlation between the parameters monitored by different channels in the temporal and spatial dimensions has become abnormal, breaking the coupling mode that should exist under normal operating conditions. This indicates that the tunneling machine has potential faults or performance degradation problems. Based on this, the failure feature vector is accurately parsed from the cross-correlation matrix. In the parsing process, firstly, based on the pre-set rule ID index in the equipment correlation judgment model, the specific channel pair data area related to the correlation coupling failure in the cross-correlation matrix can be quickly located. Then, a weighted summation aggregation algorithm is used to process the fragmented relational data of each located channel pair to obtain the failure feature vector, realizing adaptive anomaly identification of the tunneling machine. This provides comprehensive and accurate data basis for tunneling machine fault diagnosis and prediction, which helps to improve the overall reliability and operating efficiency of the tunneling machine and reduce production losses and safety risks caused by equipment failures.

[0044] Furthermore, the method for parsing the failure feature vector from the cross-relation matrix includes: extracting multiple channel pair relationship data fragments from the cross-relation matrix according to the device parameter association rules of the device association judgment model; loading the multiple channel pair relationship data fragments into the multi-task comparison engine of the device association judgment model, performing parallel spatiotemporal coupled dual-threshold decision-making, and outputting multiple failure state sets; dynamically aggregating the multiple failure state sets according to the rule ID index of the multiple channel pair relationship data fragments, and outputting the failure feature vector.

[0045] Specifically, after inputting the cross-correlation matrix into the equipment association judgment model, the matrix is ​​first sliced ​​according to the pre-set equipment parameter association rules within the model. Several channel pair relationship data fragments are extracted based on channel pairs and rule constraints. That is, focusing on the tunneling machine equipment parameter pairs, a local subset of data relationships such as correlation strength and time delay is extracted from the cross-correlation matrix. By extracting data fragments, the channel pairs that need to be analyzed can be accurately located, avoiding the computational complexity and information interference caused by uniform processing of the entire matrix, thus improving the relevance and efficiency of the analysis.

[0046] Multiple channel pair relationship data shards are input into the multi-task comparison engine of the device association judgment model, and parallel spatiotemporal coupled dual-threshold decision-making is performed. This parallel spatiotemporal coupled dual-threshold decision-making method is based on spatiotemporal correlation and dual-threshold settings. Regarding spatiotemporal correlation, the parallel spatiotemporal coupled dual-threshold decision-making comprehensively considers the temporal and spatial changes of channel data to determine whether they meet the preset association pattern. The dual thresholds set boundary conditions for normal and failed states respectively; only when the data exceeds these two threshold ranges is the channel pair determined to be in a failed state. This decision-making method can accurately identify whether each channel pair has spatiotemporal correlation coupling problems and output multiple sets of failed states, each set corresponding to the failed state information of a channel pair relationship data shard. Finally, based on the rule ID index of multiple channel pair relationship data shards, multiple sets of failed states are dynamically aggregated, and a failed feature vector is output. The rule ID index is a unique identifier assigned to each device parameter association rule, which can accurately locate the corresponding channel pair relationship data shard and its failed state set. Dynamic aggregation refers to the process of integrating various failure state sets based on rule ID indexes. A weighted summation aggregation algorithm is used to aggregate scattered failure state information into a comprehensive failure feature vector. This algorithm considers the varying importance of different rules to the overall equipment performance, assigning appropriate weights to the channel-to-data fragments corresponding to each rule ID. The weights are typically determined based on expert experience, historical data statistical analysis, and the equipment's key performance indicators. For example, the main drive motor, as a core component of the tunneling machine, will have relatively high weights for its related rules. By using weighted summation, the failure state information scattered across the channel-to-data fragments is integrated to obtain a failure feature vector. This vector contains failure state information from all channels, presenting the spatiotemporal coupling failure characteristics of the tunneling machine as a whole. This enables adaptive anomaly identification for the tunneling machine, providing intuitive and comprehensive information for fault diagnosis and maintenance decisions, helping technicians quickly locate equipment problems, and improving the reliability and operating efficiency of the tunneling machine.

[0047] Based on the failure feature vector, fault pointing probability matching is performed to output a real-time fault device group, wherein the real-time fault device group has an adaptive hierarchical fault level identifier.

[0048] Furthermore, based on the failure feature vectors, fault pointing probability matching is performed to output a real-time fault device group. The method includes: interactively obtaining multiple sets of sample feature vectors of the multiple associated devices under multiple sample fault states; constructing a three-level index tree based on the multiple associated devices, multiple sets of sample fault states, and multiple sets of sample feature vectors; loading the failure feature vectors into the three-level index tree, performing coarse screening of fault states based on pre-selected similarity to obtain a candidate fault state set; performing fine matching of the failure feature vectors and the candidate fault state set based on Euclidean distance to output multiple device fault association probability pairs; and performing device fault descending classification based on the multiple device fault association probability pairs to output the real-time fault device group, wherein the real-time fault device group has an adaptive classification fault level identifier.

[0049] Specifically, by using a human-computer interaction interface or an automated data acquisition and transmission system, characteristic data of various related devices under different preset fault states are obtained, and after data cleaning and feature extraction algorithms, multiple sets of sample feature vectors are formed. Then, a three-level index tree is constructed based on multiple related devices, multiple sets of sample fault states, and multiple sets of sample feature vectors. This three-level index tree is a hierarchical data structure for rapid retrieval. Multiple related devices serve as first-level index nodes, with each device considered an independent branch. This allows for quick location of relevant data sets for specific devices; for example, key devices such as the cutting motor, propulsion cylinder, and conveyor of a tunneling machine can be designated as first-level nodes. Next, for each device node, a second-level index is constructed based on multiple sets of sample fault states. Different preset fault states under the same device, such as overload and short circuit of the cutting motor, and abnormal pressure and leakage of the hydraulic system, are designated as child nodes under that device node. This method allows for precise identification of the data range of the device under specific fault states. Finally, under each sample fault state node, multiple sets of sample feature vectors serve as the third-level index, attaching specific vector data reflecting the device characteristics under that fault state to the corresponding fault state node. For example, feature vectors such as temperature, current, and speed corresponding to an overload fault state of the cutting motor. This hierarchical index structure significantly improves the efficiency of data retrieval and matching, thereby increasing the speed of the entire fault analysis process.

[0050] Then, the failure feature vector is loaded into the three-level index tree. Based on a pre-set similarity threshold, the similarity between the failure feature vector and the sample feature vectors in the three-level index tree is initially determined, i.e., a coarse screening of faults is performed. By calculating the similarity between the failure feature vector and the sample feature vectors under each level of the index, fault states with similarity exceeding the pre-selected threshold are filtered out to form a candidate fault state set. Based on this, the scope of fault analysis can be quickly narrowed, and obviously irrelevant fault states can be eliminated, improving the efficiency and accuracy of subsequent fine matching. After obtaining the candidate fault state set, the failure feature vector is finely matched with the fault states in the candidate fault state set based on Euclidean distance. Euclidean distance is a commonly used method to measure the distance between two vectors. In fault matching, by calculating the Euclidean distance between the failure feature vector and each sample feature vector in the candidate fault state set, the smaller the distance, the more similar the two are, that is, the greater the probability that the fault state corresponding to the sample feature vector is related to the current equipment failure. Based on the Euclidean distance calculation, the fault association probability between each candidate fault state and the corresponding equipment with the failure feature vector is obtained, forming equipment fault association probability pairs. These pairs are then sorted in descending order of probability, with higher probabilities indicating a higher likelihood of fault occurrence. A real-time fault equipment group is output. This real-time fault equipment group refers to the set of currently faulty equipment dynamically determined during tunneling machine operation, based on the matching analysis of failure feature vectors and preset fault modes. The real-time fault equipment group includes an adaptive fault level identifier, dynamically generated according to the magnitude of the equipment fault association probability and pre-defined grading rules, such as minor faults, general faults, and severe faults. This adaptive grading method can flexibly adjust according to the actual fault association situation of the equipment, more accurately reflecting the severity of equipment faults. It provides maintenance personnel with clear and intuitive equipment fault information, facilitating timely maintenance measures and improving the reliability and operating efficiency of the tunneling machine.

[0051] The historical average baseline and historical average density are dynamically updated based on the preset cumulative tunneling mileage.

[0052] Specifically, the preset cumulative tunneling mileage is a pre-defined threshold that can be set based on actual needs and expert experience. This preset cumulative tunneling mileage serves as the trigger for updates. When the tunneling machine's actual cumulative tunneling mileage reaches the preset threshold, the update process is initiated, dynamically updating the historical mean baseline and mean density. The specific update process is as follows: Key parameter data of the tunneling machine at different operating stages are continuously recorded, such as the temperature of the cutting motor and the pressure of the propulsion cylinder. When the cumulative tunneling mileage reaches the preset value, all parameter data recorded during this period are automatically extracted. Using statistical mean calculation methods, a new average value is calculated for each parameter's data, thereby updating the historical mean baseline. Simultaneously, the data for each parameter is grouped and statistically analyzed to calculate the frequency of each group's data. Then, a new density curve is plotted based on the frequency distribution, thus updating the historical mean density. Dynamic updates improve the adaptability of anomaly identification. The reference standard is automatically adjusted according to the actual use of the tunneling machine, avoiding misjudgments and omissions caused by factors such as equipment aging and changes in the working environment. At the same time, the dynamic update mechanism ensures that historical data always matches the current equipment status, providing accurate and effective reference for anomaly identification and improving the overall operational reliability and efficiency of the tunneling machine.

[0053] Based on the adaptive hierarchical fault level identifier, a fault verification priority instruction for the real-time fault equipment group is constructed, and the fault verification priority instruction is sent to the ground central control system.

[0054] Specifically, the adaptive hierarchical fault level identifier reflects the severity of each device's fault in the real-time faulty equipment group. Based on the adaptive hierarchical fault level identifier, fault verification priority instructions are constructed for devices with different fault levels, such as high priority for severe faults and low priority for minor faults. These fault verification priority instructions are then sent to the ground control system. The ground control system, located on the ground during tunneling machine operation, is the core control platform that receives these instructions and coordinates personnel and resources to sequentially verify and process the real-time faulty equipment group. This process enables the control system to arrange personnel to verify faulty equipment in an orderly manner according to priority, improving fault handling efficiency and ensuring the stable operation of the tunneling machine.

[0055] Example 2, based on the same inventive concept as the adaptive anomaly identification method for tunneling machine operating conditions in the foregoing examples, such as... Figure 2 As shown, this application provides an adaptive anomaly identification system for tunneling machine operating conditions, the system comprising: The working condition acquisition trigger module 10 is used to trigger the acquisition of the operating conditions of multiple pre-set associated equipment in the tunneling machine when a lithological abrupt change event is captured. The collaborative sampling execution module 20 is used to drive multiple distributed edge acquisition nodes of the multiple associated equipment to perform millisecond-level synchronous working condition acquisition based on pre-set collaborative sampling rules, generating multi-channel operating condition time-series data. The time-series feature backtracking module 30 is used to perform time-series feature matching backtracking on the lithological abrupt change event to locate the real-time lithological transition stage. The model retrieval module 40 is used to dynamically retrieve the equipment association judgment model based on the real-time lithological transition stage. The association coupling judgment module 50 is used to calculate the cross-correlation matrix based on the multi-channel operating condition time-series data, and then load the cross-correlation matrix into the equipment association judgment model for spatiotemporal association coupling judgment. The failure feature parsing module 60 is used to determine if the spatiotemporal association coupling judgment result is an association coupling failure, and then parse the failure feature vector from the cross-correlation matrix. The fault output module 70 is used to perform fault pointing probability matching based on the failure feature vector and output a real-time fault device group, wherein the real-time fault device group has an adaptive hierarchical fault level identifier.

[0056] The time-series feature backtracking module 30 is further configured to: use the timestamp of the lithological abrupt change event as a reference point to call the time-series data before and after a preset backtracking time window to obtain a multi-channel equipment operating condition time-series segment; output multi-modal lithological response features by solving the multi-channel equipment operating condition time-series segment, wherein the multi-modal lithological response features include the propulsion speed reduction rate, the cutterhead torque fluctuation amplitude, and the proportion of vibration signal main frequency energy; and use the multi-modal lithological response features to traverse and compare multiple standard modal feature templates of multiple standard transition stages in the lithological transition feature library to locate the real-time lithological transition stage.

[0057] The model retrieval module 40 is further configured to: interactively obtain multiple device parameter association rules for the multiple standard transition stages; after configuring multiple multi-task comparison engines for the multiple device parameter association rules, configure multiple distributed parallel processing architectures for the multiple multi-task comparison engines to complete the construction of multiple standard association judgment models; use the multiple standard transition stages as indexes to index and store the multiple standard association judgment models to generate an association judgment model library; and dynamically retrieve the device association judgment model from the association judgment model library according to the real-time lithology transition stage.

[0058] The correlation coupling decision module 50 is further configured to: combine and enumerate the multi-channel operating condition time series data to obtain multiple sets of dual-channel operating condition time series data; calculate the maximum cross-correlation number and operating condition delay of the multiple sets of dual-channel operating condition time series data based on a preset sliding segmentation window and sliding segmentation step size, and output multiple symmetric relationship matrices; aggregate the multiple symmetric relationship matrices according to the channel pair topology of the multiple sets of dual-channel operating condition time series data to generate the cross-correlation number matrix; load the cross-correlation number matrix into the equipment correlation judgment model to perform spatiotemporal correlation coupling judgment, and output the spatiotemporal correlation coupling judgment result.

[0059] The failure feature parsing module 60 is further configured to: extract multiple channel pair relationship data fragments from the cross-relationship matrix according to the device parameter association rules of the device association judgment model; load the multiple channel pair relationship data fragments into the multi-task comparison engine of the device association judgment model, perform parallel spatiotemporal coupled dual threshold decision-making, and output multiple failure state sets; dynamically aggregate the multiple failure state sets according to the rule ID index of the multiple channel pair relationship data fragments, and output the failure feature vector.

[0060] The operating condition acquisition triggering module 10 is also used to perform the following: based on the sampling window, perform sliding window standard deviation demodulation on the propulsion speed time-series signal of the tunneling machine and output the propulsion speed standard deviation; perform local extreme value density analysis on the cutterhead torque signal of the tunneling machine and output the cutterhead torque extreme value density, wherein the propulsion speed time-series signal demodulation and the cutterhead torque signal analysis are performed synchronously; when the propulsion speed standard deviation suddenly increases by more than 30% of the historical average baseline, satisfying P sampling windows, and the cutterhead torque extreme value density reaches more than Q times the historical average density, a lithological mutation event triggering command is generated; the lithological mutation event triggering command is sent to the multiple distributed edge acquisition nodes based on a unified timestamp to trigger the acquisition of the operating conditions of the multiple associated devices.

[0061] The fault output module 70 is further configured to perform the following: interactively obtain multiple sets of sample feature vectors of the multiple associated devices under multiple sets of sample fault states; construct a three-level index tree based on the multiple associated devices, multiple sets of sample fault states, and multiple sets of sample feature vectors; load the failure feature vectors into the three-level index tree, perform coarse screening of fault states based on pre-selected similarity, and obtain a candidate fault state set; perform fine matching of the failure feature vectors and the candidate fault state set based on Euclidean distance, and output multiple device fault association probability pairs; perform device fault descending classification according to the multiple device fault association probability pairs, and output the real-time fault device group, wherein the real-time fault device group has an adaptive classification fault level identifier.

[0062] The system is also used to perform: dynamic updates of the historical average baseline and historical average density based on a preset cumulative tunneling mileage.

[0063] The system is also used to execute: constructing a fault verification priority instruction for the real-time fault equipment group based on the adaptive hierarchical fault level identifier, and sending the fault verification priority instruction to the ground central control system.

[0064] In summary, any of the methods or steps described above can be stored as computer instructions or programs in various types of computer memory, and the computer instructions or programs can be recognized by various types of computer processors to implement any of the above methods or steps.

[0065] Based on the above specific embodiments of the present invention, any improvements and modifications made to the present invention by those skilled in the art without departing from the principle of the present invention shall fall within the patent protection scope of the present invention.

Claims

1. An adaptive anomaly identification method for tunneling machine operating conditions, characterized in that, The method includes: When a lithological mutation event is captured, the operating conditions of multiple pre-set associated devices in the tunneling machine are collected. Based on preset collaborative sampling rules, multiple distributed edge acquisition nodes of the multiple associated devices are driven to perform millisecond-level synchronous operating condition acquisition and generate multi-channel operating condition time-series data. The lithological abrupt change events were backtracked by time-series feature matching to locate the real-time lithological transition stage; Based on the real-time lithological transition stage, the equipment association judgment model is dynamically retrieved; After calculating the cross-correlation matrix based on the multi-channel operating condition time series data, the cross-correlation matrix is ​​loaded into the equipment association judgment model for spatiotemporal association coupling judgment. If the spatiotemporal correlation coupling determination result is a failure of correlation coupling, then the failure feature vector is parsed from the cross-correlation matrix; Based on the failure feature vector, fault pointing probability matching is performed to output a real-time fault device group, wherein the real-time fault device group has an adaptive hierarchical fault level identifier.

2. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 1, characterized in that, The method involves performing time-series feature matching and backtracking on the lithological abrupt change event to locate the real-time lithological transition stage, and includes: Using the timestamp of the lithological mutation event as a reference point, the time series data before and after the preset backtracking time window is retrieved to obtain the time series segments of the equipment operating conditions in multiple channels; By solving the time sequence segment of the multi-channel equipment operating conditions, multimodal lithological response characteristics are output, wherein the multimodal lithological response characteristics include the rate of decrease in propulsion speed, the amplitude of cutterhead torque fluctuation, and the proportion of the main frequency energy of the vibration signal; The real-time lithological transition stage is located by comparing multiple standard modal feature templates of multiple standard transition stages in the lithological transition feature library with the multimodal lithological response features.

3. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 2, characterized in that, Based on the real-time lithological transition stage, the method of dynamically retrieving the equipment association judgment model includes: Interactively obtain the association rules of multiple device parameters for the multiple standard transition stages; After configuring multiple multi-task comparison engines for the association rules of the multiple device parameters, multiple distributed parallel processing architectures of the multiple multi-task comparison engines are configured to complete the construction of multiple standard association judgment models. The multiple standard transition stages are used as indexes to index and store the multiple standard association judgment models, thereby generating an association judgment model library. Based on the real-time lithological transition stage, the equipment association judgment model is dynamically retrieved from the association judgment model library.

4. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 3, characterized in that, After calculating the cross-correlation matrix based on the multi-channel operating condition time-series data, the cross-correlation matrix is ​​loaded into the equipment association judgment model for spatiotemporal association coupling determination. The method includes: By combining and enumerating the multi-channel operating condition time series data, multiple sets of dual-channel operating condition time series data are obtained. Using a preset sliding segmentation window and sliding segmentation step size as the calculation basis, the maximum cross-correlation coefficient and operating delay of the multiple sets of dual-channel operating condition time series data are calculated, and multiple symmetric relation matrices are output. Based on the channel pair topology of the multiple sets of dual-channel operating condition time series data, the multiple symmetric relationship matrices are aggregated to generate the cross-relationship matrix; The cross-correlation matrix is ​​loaded into the device association judgment model to perform spatiotemporal association coupling judgment, and the spatiotemporal association coupling judgment result is output.

5. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 4, characterized in that, The method for resolving failure feature vectors from the cross-correlation matrix includes: Based on the device parameter association rules of the device association judgment model, multiple channel pairs of relationship data are extracted from the mutual relationship matrix for fragmentation; The multiple channel relationship data are sharded and loaded into the multi-task comparison engine of the device association judgment model, and parallel spatiotemporal coupled dual threshold decision is executed to output multiple failure state sets. Based on the rule ID index of the multiple channels for sharding relational data, the multiple failure state sets are dynamically aggregated, and the failure feature vector is output.

6. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 1, characterized in that, When a lithological mutation event is captured, the operating conditions of multiple pre-set associated devices in the tunneling machine are collected. The method includes: Based on the sampling window, the advance speed time sequence signal of the tunneling machine is demodulated using a sliding window standard deviation, and the advance speed standard deviation is output. Local extreme value density analysis is performed on the cutterhead torque signal of the tunneling machine to output the cutterhead torque extreme value density. The demodulation of the propulsion speed timing signal and the analysis of the cutterhead torque signal are performed simultaneously. When the standard deviation of the propulsion speed suddenly increases by more than 30% of the historical mean baseline, satisfying P sampling windows, and the extreme density of the cutterhead torque reaches more than Q times the historical mean density, a lithological mutation event trigger command is generated. The lithological mutation event trigger command is sent to the multiple distributed edge acquisition nodes based on a unified timestamp, triggering the acquisition of the operating conditions of the multiple associated devices.

7. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 1, characterized in that, The method includes performing fault pointing probability matching based on the failure feature vector and outputting a real-time faulty device group. Interactively obtain multiple sets of sample feature vectors of the multiple associated devices under multiple sets of sample fault states; Based on the multiple associated devices, multiple sets of sample fault states, and multiple sets of sample feature vectors, a three-level index tree is constructed. The failure feature vector is loaded into the three-level index tree, and the failure state is coarsely screened based on the pre-selected similarity to obtain a candidate failure state set. Based on Euclidean distance, the failure feature vector is precisely matched with the failure state of the candidate failure state set, and multiple equipment failure association probability pairs are output. Based on the multiple device fault association probabilities, the device faults are classified in descending order, and the real-time fault device group is output, wherein the real-time fault device group is equipped with an adaptive classification fault level identifier.

8. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 6, characterized in that, The historical average baseline and historical average density are dynamically updated based on the preset cumulative tunneling mileage.

9. The adaptive anomaly identification method for tunneling machine operating conditions as described in claim 1, characterized in that, Based on the adaptive hierarchical fault level identifier, a fault verification priority instruction for the real-time fault equipment group is constructed, and the fault verification priority instruction is sent to the ground central control system.

10. An adaptive anomaly identification system for tunneling machine operating conditions, characterized in that, The steps for implementing the method according to any one of claims 1 to 9 include: The working condition acquisition trigger module (10) is used to trigger the acquisition of the operating conditions of multiple pre-set associated equipment in the tunneling machine when a lithological change event is captured; The collaborative sampling execution module (20) is used to drive multiple distributed edge acquisition nodes of the multiple associated devices to perform millisecond-level synchronous operating condition acquisition based on preset collaborative sampling rules, and generate multi-channel operating condition time series data; The time-series feature backtracking module (30) is used to perform time-series feature matching backtracking on the lithological abrupt events and locate the real-time lithological transition stage; The model retrieval module (40) is used to dynamically retrieve the equipment association judgment model based on the real-time lithological transition stage; The correlation coupling decision module (50) is used to calculate the cross-correlation matrix based on the multi-channel operating condition time series data, and then load the cross-correlation matrix into the equipment correlation decision model to make a spatiotemporal correlation coupling decision. The failure feature analysis module (60) is used to determine whether the spatiotemporal correlation coupling judgment result is a correlation coupling failure, and then to parse the failure feature vector from the cross-correlation matrix. The fault output module (70) is used to perform fault pointing probability matching based on the failure feature vector and output a real-time fault device group, wherein the real-time fault device group has an adaptive hierarchical fault level identifier.

Images