Small-hole-diameter TBM tunnel supporting measure design method

Abstract

The invention relates to the technical field of tunneling engineering, in particular to a small-hole-diameter TBM tunnel supporting measure design method which comprises the steps that sensors are arranged at key positions of a TBM to collect strain signals and obtain material parameters, and target interference signals are separated from mixed signals through a component kinematic model; analyzing the signal features to establish a quantitative fingerprint database and generate a real-time load spectrum; integrating a matching result, a material parameter and a load spectrum, performing fusion calculation on a risk situation index, and identifying a dominant interference mode; inquiring an adaptive strategy matrix so as to dynamically execute hierarchical regulation and control; and finally, a strategy matrix is optimized through regulation and control effect feedback, subsequent tasks are planned, a closed-loop process of sensing, analysis, decision making, regulation and control and optimization is formed, and dynamic intelligent design from passive response to active predictive support is achieved.

Description

Technical Field

[0001] This invention relates to the field of tunnel boring engineering technology, and more specifically, to a design method for support measures in small-diameter TBM tunnels. Background Technology

[0002] In the existing support design of small-diameter TBM tunnels, a particularly hidden and extremely dangerous technical problem is "ghost interference" and the hidden damage chain it causes. Within the extremely limited space of the tunnel boring machine, although a safety clearance is reserved on the static design drawings between the support equipment (such as the bolt drilling arm) and the key structural components of the TBM body (such as the propulsion cylinder support and pipeline cable tray), this safety boundary may be quietly breached under actual complex dynamic working conditions.

[0003] The dynamic failure scenario manifests as follows: during actual tunneling, the TBM will generate continuous "snake-like" swaying and attitude fine-tuning in order to correct or adapt to the curve; at the same time, various equipment also have inherent vibrations during operation. This complex dynamic motion makes the actual relative motion envelope between components far exceed the static design range, which may lead to low-frequency, high-stress periodic "scratching" or "collision".

[0004] The fundamental challenge that existing technological approaches face in addressing this problem is: Staticization of design paradigms: Existing support designs mainly rely on static gap checks, which cannot characterize or predict the above-mentioned dynamic composite motions. As a result, "ghost interference" cannot be detected at the drawing stage and is difficult to reproduce during no-load commissioning.

[0005] Insufficient monitoring and diagnostic capabilities: Traditional vibration monitoring technology struggles to accurately separate and identify the unique stress wave signal induced by such weak, intermittent contact from strong background noise, making it impossible to achieve early warning and precise location of interference events.

[0006] Disconnect between damage assessment and control: Even when abnormal vibrations are detected, existing methods lack models that can quantitatively correlate transient signals with the long-term fatigue damage accumulation of critical structural components in real time. This leads to delayed maintenance decisions and makes it impossible to intelligently coordinate and control with tunneling process parameters (such as thrust and torque) to proactively suppress interference.

[0007] Lack of adaptive and learning mechanisms: Faced with constantly changing geological conditions and equipment status during the tunneling process, fixed support and maintenance strategies cannot learn and optimize from historical intervention events and control effects, and the system does not have the ability to continuously evolve to cope with new risks.

[0008] The danger of this "ghost interference" lies in its insidious and cumulative nature. It typically doesn't cause immediate functional failure, but like the ghost of "metal fatigue," it continuously applies cyclical stress during hundreds of kilometers of tunneling, potentially leading to hidden cracks or loosening of connections in critical structural components (such as propulsion cylinder lugs). Under extreme conditions (such as high-thrust tunneling in hard rock), it can induce sudden structural failure, causing TBM jamming or shield damage, resulting in significant safety risks and economic losses, and subsequent repairs in confined spaces are extremely difficult.

[0009] Therefore, there is an urgent need in this field to develop an innovative design method that can fundamentally address the dynamic and hidden risk of "ghost interference" and realize a paradigm shift from static verification to dynamic perception, from delayed response to real-time assessment and prediction, and from fixed strategies to adaptive intelligent control, so as to ensure the safe and efficient long-distance tunneling of small-diameter TBMs. Summary of the Invention

[0010] To address the shortcomings of existing technologies, the present invention aims to provide a design method for support measures in small-diameter TBM tunnels.

[0011] To achieve the above objectives, the present invention provides the following technical solution: A design method for support measures in small-diameter TBM tunnels includes the following steps: S1: Distributed strain wave sensors are installed on the base of key structural components and support equipment of the tunnel boring machine to collect multi-source mixed strain wave signals and obtain the pre-set material fatigue characteristic parameters of the key structural components. S2: Based on a predefined component kinematic model, the target interference signal associated with the potential interference component pair is separated from the multi-source mixed strain wave signal; S3: Perform time-frequency domain analysis on the target interference signal, extract and quantize its characteristic parameters to generate the real-time load spectrum of key structural components, and establish a quantitative interference fingerprint library for each potential interference component based on the quantization results; S4: Based on the matching results of the target interferometric signal and the quantitative interferometric fingerprint database, and combined with the preset material fatigue characteristic parameters and real-time load spectrum, the fusion risk situation index is generated by fusion calculation, and the current dominant interferometric mode type is determined. S5: Based on the integrated risk situation index and the type of intervention mode, query the preset adaptive control strategy matrix, dynamically determine and execute the hierarchical control instructions and their control parameters; S6: Use the execution effect data of the hierarchical control command as feedback to update the parameters of the corresponding decision logic unit in the adaptive control strategy matrix, and dynamically optimize the timing of support control and maintenance window of subsequent tunneling cycles based on the updated model.

[0012] Furthermore, the acquisition of multi-source hybrid strain wave signals in step S1 specifically includes: Fiber grating sensors or piezoelectric sensors are used as distributed strain wave sensors; Each distributed strain wave sensor is rigidly connected and placed in the stress concentration area of ​​key structural components and support equipment base. Synchronous data acquisition is performed on all distributed strain wave sensors at a sampling frequency of not less than 10kHz; Each distributed strain wave sensor acquires a strain wave signal and assigns a corresponding spatial location code to form a multi-source hybrid strain wave signal.

[0013] Furthermore, step S2, separating the target interference signal, specifically includes: The expected relative kinematic characteristics of each potential interfering component pair are calculated based on the component kinematic model; The blind source separation algorithm was used to process the multi-source mixed strain wave signal, and several independent source signals were initially separated. The time-frequency characteristics of independent source signals are matched with the expected relative kinematic characteristics. Independent source signals with a matching degree exceeding the preset interference feature matching degree threshold are identified and extracted as target interference signals associated with specific potential interference component pairs.

[0014] Furthermore, potential interference components are identified in the following way: Based on the digital prototype model of the tunnel boring machine and support equipment, identify all component combinations with relative motion; For each component assembly, a parameterized multibody dynamics numerical simulation model is established based on the kinematic constraints determined by its actual mechanical connection method. Using a parameterized multibody dynamics numerical simulation model, the convex hull boundary of the spatial point cloud set formed by the motion trajectories of all feature points on the surface of each moving part during its full working cycle is calculated. The component pairs that have static design gaps between components obtained from the analysis of the digital prototype model that are less than the safety threshold and whose calculated convex hull boundaries intersect are predefined as potential interference component pairs, and their initial interference risk profiles are established.

[0015] Furthermore, step S3, establishing the quantitative interference fingerprint database, specifically includes: The quantization results of the feature parameters are bound to the corresponding potential interferometer pairs; A relational database is established as a quantitative interference fingerprint library, in which each record represents a unique interference pattern of a potential pair of interfering components; Each record contains the following data fields: unique identifier for the component, center frequency and bandwidth of the characteristic frequency band, time-domain statistical characteristics of the transient impact waveform, and damping ratio and mode shape vector of the signal mode. The initial profile of the interference risk of each potential interference component pair is associated with its record in the quantitative interference fingerprint database.

[0016] Furthermore, the specific process of generating the fusion risk situation index in step S4 includes: Three sub-matching degrees are extracted: characteristic frequency band energy ratio, transient impact waveform correlation, and signal mode confidence. The three sub-matching degrees are multiplied by preset contribution weight coefficients and then linearly weighted and summed to obtain the initial risk contribution value. The initial risk contribution value is then mapped to the zero-to-one interval through a normalization function to obtain the real-time risk value of component interference. Rainflow counting statistics are performed on the load amplitude sequence in the real-time load spectrum to obtain the load cycle distribution; the load cycle distribution is convolved with the stress-life curve in the preset material fatigue characteristic parameters to calculate the equivalent fatigue damage degree under the current load cycle; the equivalent fatigue damage degree is added to a historical damage memory factor, and the sum is then limited to the range of zero to one to obtain the material damage cumulative risk value; where the historical damage memory factor is equal to the material damage cumulative risk value of the previous calculation cycle multiplied by a preset historical attenuation coefficient less than one; The real-time risk value of component interference, the cumulative risk value of material damage, and a differential term reflecting the trend of risk change are fused together in a multi-dimensional situational analysis to obtain the final fused risk situational index.

[0017] Furthermore, the multi-dimensional situational fusion calculation is performed according to the following steps: Step 1: Calculate the geometric mean of the real-time risk value of component interference and the cumulative risk value of material damage; Step 2: Squaring the geometric mean yields the basic situational components; Step 3: Calculate the cube root of the absolute value of the differential term of the risk change trend to obtain the trend change component. Step 4: Add the basic situational component to the trend change component to obtain the unweighted preliminary situational value; Step 5: Multiply the preliminary situation value by the adaptive weighting coefficient to obtain the final fused risk situation index.

[0018] Furthermore, step S4, determining the currently dominant interferometry mode type, specifically includes: For each potential pair of interfering components, calculate the multidimensional feature similarity between the feature parameters stored in its quantized interferometric fingerprint database and the corresponding feature parameters of the target interferometric signal; By coupling the multidimensional feature similarity with a weighting factor that reflects the proportion of energy of the signal components in the target interference signal, a comprehensive matching degree is obtained. Set an interference mode confidence threshold and compare the overall matching degree of all potential interference component pairs with this confidence threshold; If there is a unique overall match degree that exceeds the confidence threshold, then the interference mode of the potential interfering component pair corresponding to that value is determined as the currently dominant interference mode type. If there are multiple combined matching degrees that exceed the confidence threshold, the interference mode of the potential interfering component pair corresponding to the one with the highest value will be determined as the currently dominant interference mode type.

[0019] Furthermore, step S5, which involves dynamically determining and executing tiered control commands and their control parameters, specifically includes: Query the preset adaptive control strategy matrix to find the strategy unit corresponding to the current fusion risk situation index and intervention mode type, and read one or more control instruction identifiers to be executed stored therein; Based on the preset risk level range of the integrated risk situation index, a set of specific control parameter values ​​corresponding to each control instruction identifier are dynamically calculated from the strategy unit. According to the preset execution logic, the control command identifier and its corresponding specific control parameter value are synchronously or sequentially sent to the bottom controller of the tunnel boring machine and support equipment to drive the corresponding physical actuators to complete the control action.

[0020] Compared with the prior art, the present invention has the following beneficial effects: This invention, by introducing a dynamic sensing framework that combines a distributed strain wave sensing network with component kinematic models, completely changes the traditional design paradigm that relies on static gap verification. By combining the dynamic envelope and relative motion trajectory described by the predefined component kinematic models, the system can accurately separate the often extremely weak target interference signals associated with specific potential interference component pairs (such as the anchor drill arm and the propulsion cylinder support) from the comprehensive signals containing multiple vibration sources such as cutterhead rock breaking and equipment operation, using advanced algorithms such as blind source separation. This capability enables the early identification and precise positioning of "ghost interference"—that is, intermittent low-energy contact caused by composite dynamic motions such as TBM correction swaying and equipment vibration—for the first time. It solves the fundamental defect of existing technologies that cannot detect such hidden risks before accidents occur due to crude sensing methods, and provides a crucial time window for active protection. The system not only identifies interference patterns but also, through time-frequency domain analysis of the target interference signal, retrieves the real-time load spectrum acting on key structural components. Furthermore, this load spectrum is combined with pre-set material fatigue characteristic parameters (such as the SN curve of a specific component) to quantify its fatigue damage contribution using a cumulative damage model. Simultaneously, the matching degree of the interference signal is normalized to obtain an index characterizing the instantaneous severity. Finally, through multi-dimensional situational fusion calculations, instantaneous risk, cumulative damage, and risk change trends are integrated to generate a unified, quantitative fusion risk situation index. Based on this index and the identified specific interference pattern, the system queries a pre-set adaptive control strategy matrix, dynamically calculates, and triggers one or more graded control commands, such as anti-phase acoustic vibration control, stiffness adjustment, and propulsion waveform adjustment. This proactive control, based on quantitative risk assessment and coordinated with the tunneling process depth, can effectively interrupt the accumulation process of periodic micro-damage, preventing the initiation and development of latent cracks in key structural components from the root, and shifting risk management from post-event remediation to in-event suppression. This invention further establishes a closed-loop optimization mechanism with self-learning and continuous evolution capabilities, endowing the support system with unprecedented long-term adaptability and intelligence. The system uses the execution effect of each graded control command (such as the risk index reduction rate and load peak attenuation) as feedback signals to drive iterative optimization of the corresponding decision logic units in the adaptive control strategy matrix, such as adjusting command priorities or micro-control parameter calculation rules. Simultaneously, this real-time data is used to update the component kinematic model, the quantified interference fingerprint database, and the damage prediction model, making them increasingly accurate in reflecting the dynamic characteristics and damage patterns of the actual equipment. Based on the updated models and strategies, the system can proactively and dynamically optimize the support control sequence and preventative maintenance windows for subsequent tunneling cycles. This means that the system can not only cope with currently identified interferences but also predict and prevent future risks that may arise under new working conditions by learning from historical data, thereby achieving a paradigm shift from "fixed program response" to "autonomous intelligent evolution," significantly improving the overall robustness of small-diameter TBMs in safe, efficient, and reliable operation under long-distance, complex geological conditions. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the system composition and data flow of the small-diameter TBM tunnel support design method of the present invention; Figure 2 This is a schematic diagram of the process of separating the target interference signal from a multi-source mixed signal and establishing a quantization fingerprint library in this invention; Figure 3 This is a schematic diagram illustrating the logic of generating a fusion risk situation index and determining the interference mode in this invention. Detailed Implementation

[0022] Reference Figures 1 to 3 A design method for support measures in small-diameter TBM tunnels, comprising: S1: Distributed strain wave sensors are deployed on the bases of key structural components of the tunnel boring machine and support equipment to collect multi-source hybrid strain wave signals and obtain pre-set material fatigue characteristic parameters of the key structural components. This step is the data sensing and input foundation for the design method described above. By deploying a distributed sensor network at key locations where the tunnel boring machine and support equipment interact, wide-area, synchronous, and high-precision capture of structural micro-strain waves directly induced by the dynamic process of "ghost interference" is achieved for the first time, providing the original "multi-source hybrid strain wave signal" as a data source for subsequent analysis. Simultaneously, "obtaining pre-set material fatigue characteristic parameters of the key structural components" refers to introducing known physical performance data (such as SN curves) related to the structural materials themselves. This provides indispensable input conditions reflecting the inherent properties of the materials for establishing a quantitative correlation between dynamic signals and long-term structural damage. This step together constitutes the physical world information entry point for all subsequent intelligent analysis, evaluation, and decision-making. In one specific implementation, the process of acquiring multi-source hybrid strain wave signals in step S1 is as follows: Fiber Bragg grating sensors are selected as distributed strain wave sensors because they possess anti-electromagnetic interference and long-term stability, making them suitable for the complex electromagnetic environment within tunnels. Specifically, based on the finite element static analysis results of key structural components of the tunnel boring machine (such as the propulsion cylinder lugs and main bearing flanges) and support equipment (such as the rotary base of the anchor drilling rig), stress concentration areas (such as weld transition zones and around bolt holes) under typical loads are identified as the target areas for sensor deployment. High-strength epoxy resin adhesive, combined with specialized clamps, rigidly connects each fiber Bragg grating sensor to the center of the target area on the cleaned and polished metal surface, ensuring minimal transmission path loss of the strain wave within the structure. After installation, a protective cover is used for physical protection. The data acquisition system was configured by connecting all installed fiber Bragg grating sensors to a multi-channel high-speed demodulator. A uniform sampling frequency of 20kHz was set, sufficient to capture the high-frequency stress wave components generated by transient collisions between components (typically with energy distribution in the 1kHz-8kHz range). Hardware trigger signals ensured strict synchronization of the acquisition start time for all channels, with synchronization errors controlled within microseconds. At the software level, each acquisition channel's data stream was assigned a unique spatial location code identifier, corresponding one-to-one with the sensor's physical installation location (e.g., "left top shield - No. 3 propulsion cylinder lug - longitudinal"). The synchronized data streams from all channels, arranged in the coded order, collectively constituted the original dataset of multi-source hybrid strain wave signals with a clear spatiotemporal correlation, providing a traceable data foundation for subsequent signal separation. In one specific implementation, the acquisition of pre-set material fatigue characteristic parameters for key structural components is achieved through a process combining material traceability, customized experiments, and engineering modeling. Specifically, standard fatigue test blanks are cut from spare parts or materials of the same batch from key structural components (e.g., lug forgings for propulsion cylinders) planned for monitoring in the tunnel boring machine (TBM). These blanks are then precision-machined into standard fatigue specimens conforming to national standards. The correspondence between these specimens and actual machine parts in terms of heat treatment batches and forging processes is recorded to ensure the traceability of material performance. Customized fatigue performance tests are performed for actual working conditions. Instead of directly referencing standard data from material handbooks, the typical stress state of the key structural component during TBM operation is simulated on a testing machine. For example, an asymmetric cyclic stress ratio is set based on historical load statistics, and a loading rate similar to the operating pulse frequency of the TBM is applied. Stress-life curves and cyclic stress-strain curves of the specific material under application conditions are obtained through group testing. A damage correlation model from material specimens to actual structures is established. Finite element analysis is used to calculate the local stress concentration factor of key structural components under typical design loads. Combined with the material cyclic performance data obtained from the aforementioned experiments, the fatigue life data of the material specimens is converted and corrected into component-level fatigue resistance expressions corresponding to specific structural geometry and stress gradients using the critical domain method or energy method. A structured pre-set database of material fatigue characteristic parameters is constructed. The fatigue performance data obtained above, associated with specific components, specific process batches, and specific operating conditions, including but not limited to stress-life curve parameters, fatigue limits, crack propagation thresholds, and Paris law coefficients, are encoded and stored as a structured parameter set indexed by the unique component identifier, which can be directly called in subsequent steps. This completes the conversion and preparation from general material data to high-confidence engineering data specifically for this design method. In one specific implementation, taking the propulsion cylinder lug (made of low-alloy high-strength steel 34CrNiMo6), a crucial load-bearing component of the tunnel boring machine's propulsion system, as an example, the process of establishing a damage correlation model from material samples to the actual structure is illustrated. Based on standard samples taken from forgings of the same batch as the lug, the high-pulse load characteristics during tunnel boring (e.g., stress ratio R=0.1, loading frequency of 5Hz) were simulated on a fatigue testing machine. The basic stress-life (SN) curve of the material under the target working condition was obtained, and its expression is as follows: Where C and m are material constants. A high-fidelity three-dimensional finite element parametric model of the propulsion cylinder lug is established. Static analysis is performed based on its design working load (e.g., maximum thrust 25,000 kN) and constraints to accurately calculate the local stress concentration factor at the known critical point, the circular arc transition zone at the root of the lug. The stress gradient distribution cloud map of the nodes in this region is extracted. The point method (Theory of Critical Distances) from the critical domain method is applied for correlation transformation: based on the material's fatigue limit... and fatigue strength reduction coefficient (Depend on Based on the material's sensitivity coefficient q to notches, a characteristic critical distance parameter L applicable to this specific geometry and stress field is determined. Using this critical distance L, the stress field distribution near the critical point of the lug calculated by the finite element model is equivalently correlated with the stress amplitude parameter in the material's SN curve. Through integration or averaging, the fatigue life data at the material specimen level (in terms of...) is then... (As a variable) is transformed and corrected to the "equivalent structural stress amplitude" at the local hot spot of the ear seat. The expression for component-level fatigue resistance, where variables are variables, may be modified to the following form: ,in and This includes specific parameters that incorporate the geometry, notch effect, and stress gradient influence of the lug. This expression is stored as a core part of the pre-set material fatigue characteristic parameters for this particular key structural component (propeller cylinder lug), enabling subsequent steps to utilize the real-time load spectrum (corresponding to...) When performing fatigue damage accumulation calculations, it can directly and accurately reflect the true fatigue resistance of a specific component, rather than generalized material handbook data.

[0023] S2: Based on a predefined component kinematic model, the target interference signal associated with potential interfering component pairs is separated from the multi-source mixed strain wave signal. This step is crucial for signal preprocessing and target information extraction. Since the acquired "multi-source mixed strain wave signal" is a comprehensive reflection of all vibration sources (such as cutterhead rock breaking, equipment operation, and structural vibration) of the tunnel boring machine under complex tunneling conditions, directly identifying the weak signal caused by interference from specific components is extremely difficult. This step creatively introduces a "predefined component kinematic model," which contains the known geometric constraints and relative motion laws between each "potential interfering component pair" (such as the drill arm and cylinder support). Based on this model, the mixed signal is decoupled and blind source separated, effectively filtering out background noise and irrelevant signal interference, accurately "separating" the "target interference signal" directly corresponding to the specific interference physical process, thus transforming the abstract mixed data stream into a characteristic signal representing a specific interference event that can be specifically analyzed in subsequent steps. In one specific implementation, the specific process of separating the target interference signal in step S2 is as follows: Based on a predefined component kinematic model, for each identified potential interference component pair (e.g., the anchor drill arm and the adjacent propulsion cylinder support), the relative displacement vector and angular velocity vector changing with time are calculated, and considering the connection gap and manufacturing tolerance, the theoretical relative motion trajectory of the component pair under all possible contact or proximity states within a complete working cycle is simulated and generated. Then, the characteristic frequency components, expected impact intervals, and motion direction characteristics in the theoretical trajectory are extracted by Fourier transform, and this information is quantified into a set of expected relative kinematic characteristics unique to the component pair. The multi-source mixed strain wave signal with spatiotemporal identification obtained in step S1 is processed by a fast independent component analysis algorithm based on maximizing negative entropy. This algorithm takes the synchronous signals from all distributed strain wave sensors as the observation input, and iteratively optimizes the separation matrix to find several statistically most independent source signal components, thereby initially separating multiple independent source signals that are independent in the time domain. For each initially separated independent source signal, a Hilbert-Huang transform is performed to obtain its instantaneous frequency and marginal spectrum, and its zero-crossing rate, short-time energy, and other time-frequency domain characteristics are calculated. These actual time-frequency characteristics extracted from the independent source signals are then matched one-by-one with the calculated expected relative kinematic characteristics of each potential interfering component pair. The matching process calculates the coherence function values ​​and correlation coefficients of the impact intervals within the characteristic frequency bands, and comprehensively obtains a numerical matching degree between 0 and 1. An interference feature matching degree threshold of 0.75 is set (the threshold is mainly based on historical data statistics and simulation verification). Through feature extraction and matching degree calculation of independent source signals under a large number of known working conditions (including normal tunneling, known slight interference, and no interference states), statistical analysis shows that the matching degree distribution of background vibration and non-interference signals is usually below 0.6, while the matching degree of signals confirming interference is generally higher than 0.8. To ensure identification reliability while avoiding false alarms, 0.75 is selected as the threshold, which is in a significant dividing interval between the two types of sample distributions, and after... Closed-loop simulation testing in the digital twin model effectively balances detection sensitivity and specificity, ensuring high confidence in judging real interference events under complex conditions. The matching degree of all independent source signals is compared with this threshold. Independent source signals with a matching degree exceeding 0.75 are determined to be generated by the dynamic interaction of the specific potential interference component pair and are accurately extracted from the mixed signal as the target interference signal uniquely associated with that component pair for subsequent in-depth analysis. Signals with a matching degree below the threshold are classified as background vibrations or other irrelevant interference. In a specific implementation example, a fast independent component analysis algorithm based on maximizing negative entropy is used to process the mixed signal synchronously acquired by eight strain wave sensors. The input eight-channel signal data undergoes centering and whitening preprocessing to remove correlations. The separation matrix is ​​iteratively optimized with the objective function of maximizing the sum of the negative entropies of each output component, and fast convergence is achieved using a fixed-point iterative approach approximating Newton's method. The algorithm separates four statistically independent source signal components from the mixed signal. Verification shows that the main energy of one component is concentrated between 120Hz and 350Hz, which highly matches the expected characteristic frequencies of the potential interfering component pair of the anchor drill arm and the propulsion cylinder support, thus successfully extracting it as the target interfering signal. In a specific example, the synchronous signals collected by 16 distributed strain wave sensors are used to form a 16-dimensional observation vector. After centering and whitening preprocessing, the vector is input into a fast independent component analysis algorithm based on maximizing negative entropy. This algorithm iteratively updates a 16×16 separation matrix at fixed points, using the sum of the negative entropies of the output signal components as the independence criterion. After several iterations and convergence, the algorithm successfully decomposes the 16-dimensional mixed signal into 16 statistically independent source signal components. Among them, the 7th output component exhibits a clear periodic impact pattern in the time domain. Its instantaneous frequency characteristics are highly consistent with the expected kinematic model of the potential interference pair of components, namely the rotating base of the anchor drilling rig and the inner surface of the shield, and are therefore initially identified as a valuable independent source signal for subsequent feature matching. In a specific example, a Hilbert-Huang transform was performed on a separated independent source signal, revealing that its instantaneous frequency fluctuates stably within the range of 240-260Hz, with a significant peak in the marginal spectrum at 255Hz. Simultaneously, its zero-crossing rate was calculated to be 85 times per second, and its short-time energy exhibits a periodic pulse characteristic of once every 0.8 seconds. The actual time-frequency characteristics of this signal were matched with the expected relative kinematic characteristics of the potential interfering component pair "anchor drill arm-propeller cylinder support": the expected characteristic frequency is 245Hz, and the theoretical impact interval is 0.82 seconds. The two match highly in characteristic frequency and energy pulse period, with a calculated comprehensive matching degree of 0.91, far exceeding the preset threshold. Therefore, the signal was successfully identified and extracted as the target interfering signal associated with this component pair. In one specific implementation, the process of identifying potential interfering component pairs is completed through a refined workflow combining digital prototyping, multibody dynamics simulation, and computational geometry analysis, as follows: Based on a complete 3D digital prototype model including the tunnel boring machine (TBM) main unit and a full set of support equipment, the assembly constraint relationships and kinematic pair information defined in the model tree are used to automatically traverse and filter all component combinations with direct or indirect relative motion. For example, in addition to the obvious combination of the anchor drilling rig arm and the inner wall of the shield, it is also necessary to identify non-obvious combinations such as the slewing bearing of the segment assembly machine and the track beam of the rear material hoist, which may have motion intersections under specific postures. For each selected component combination, based on the kinematic constraints determined by its actual mechanical connection method (such as hinge, slide rail, fixed connection), its corresponding parametric numerical simulation model is established in the multibody dynamics simulation software environment. Taking the anchor drilling rig arm as an example, its model will accurately include the extension and pitch kinematic pairs of each arm section driven by hydraulic cylinders, the clearance parameters at each joint, and the drive speed curve. Next, the parameterized multibody dynamics numerical simulation model is driven to simulate all possible movements of the components throughout a full working cycle (such as the entire process of a bolt drilling rig from retraction, positioning, drilling, to retraction), and the trajectories of pre-arranged mesh feature points on the surface of the moving components in space are recorded. All trajectory points are aggregated into a three-dimensional spatial point cloud set, and the minimum convex hull boundary of this point cloud set is calculated using a fast convex hull algorithm. This convex hull boundary represents the maximum spatial range that the moving component may occupy after considering all dynamic poses. Interference probability determination and documentation are performed: on the one hand, the static design gap of the component assembly in the nominal zero position is directly parsed from the digital prototype model; on the other hand, the calculated dynamic convex hull boundaries of each moving component are subjected to Boolean operations to check whether there is any intersection. Combinations of components with static design clearances smaller than a preset safety threshold (this threshold is typically determined based on recommended minimum safe clearances for moving parts in mechanical design manuals, combined with field maintenance experience data from similar tunnel boring machines; for example, it is generally set to 15 mm to accommodate vibration, deformation, and tolerance accumulation) and overlapping dynamic convex hull boundaries are formally predefined as potential interference component pairs. An initial interference risk profile is created for this pair of components, recording its static clearance value, dynamic convex hull intersection volume, typical operating conditions where the intersection occurs, and information on the key structural components involved. This provides priority guidance for subsequent quantitative fingerprint database establishment and real-time monitoring. S3: Perform time-frequency domain analysis on the target interference signal, extract and quantify its characteristic parameters to generate the real-time load spectrum of key structural components, and establish a quantized interference fingerprint database for each potential interference component pair based on the quantization results. This step undertakes the dual function of converting time-domain signals into quantifiable engineering parameters and a knowledge base. First, through in-depth time-frequency domain joint analysis of the "target interference signal" (such as wavelet transform, Hilbert-Huang transform, etc.), not only can "characteristic parameters" representing interference impact intensity, frequency characteristics, and mode shapes be extracted, but also the "real-time load spectrum" acting on the key structural components can be generated in reverse "decoupling," realizing the tracing from vibration phenomena to mechanical excitation. Second, the quantified characteristic parameters are systematically associated and stored with specific "potential interference component pairs," thereby "establishing a quantized interference fingerprint database." This fingerprint database is essentially a continuously enriched database that digitizes and characterizes each specific interference mode (such as component A and component B scraping at a certain frequency and waveform), providing comparison standards and a knowledge base for subsequent rapid pattern recognition and risk assessment. In one specific implementation, the process of performing time-frequency domain analysis on the target interferometric signal to generate the real-time load spectrum of key structural components is achieved through a refined process based on dynamic inversion and signal decoupling, as follows: Dynamic model calibration and frequency response function acquisition are performed. For the key structural component of interest (e.g., the lug of a propulsion cylinder), a parameterized frequency response function matrix is ​​established between the target interferometric signal sensor deployment point and the possible load application point, combining its high-fidelity finite element model and measured modal parameters. This matrix characterizes the transfer characteristics from dynamic input load to output strain response and is verified and corrected in the key frequency band (e.g., 0-1000Hz) through hammer impact experiments. High-resolution time-frequency joint analysis is performed on the target interferometric signal (i.e., the separated strain wave signal associated with a specific potential interferometric component pair). Methods such as synchronous compressed wavelet transform or generalized S-transform are used to obtain the energy distribution representation of the signal in the time-frequency domain, i.e., the wavelet coefficient matrix or time-frequency spectrum, thereby simultaneously preserving the temporal and frequency component information of the load event. Load time history reconstruction based on inverse frequency response function operation is performed. The time-frequency representation of the obtained target interference signal is deconvolved with the obtained frequency response function of the corresponding sensing path in the corresponding frequency band. In the frequency domain, this process is equivalent to dividing the response spectrum by the frequency response function and converting it back to the time domain, thereby decoupling and obtaining the estimated time history of the dynamic input load acting on the key structural component corresponding to the target interference event. For example, the sequence of dynamic lateral force acting on the lug connecting pin as a function of time is calculated. Finally, engineering-based cyclic statistics are performed on the reconstructed dynamic load time history. Using the rainflow counting algorithm or range-pair method, all complete stress cycles in the load time history are identified and extracted, and their amplitude, mean, and occurrence frequency are counted. Finally, a real-time load spectrum of the key structural component corresponding to the current analysis time window is formed, with the load amplitude on the x-axis and the number of cycles on the y-axis, which can be directly used for fatigue damage assessment. For example, this process can be used to calculate the dynamic impact load spectrum acting on the lug from the strain wave signal of the lug caused by interference of a section of anchor drilling rig arm. The typical load amplitude range may be between 50kN and 200kN, and the main cycle frequency is consistent with the interference characteristic frequency. In one specific implementation, the process of establishing a quantitative interferometric fingerprint database is completed through a systematic data archiving and knowledge-based workflow, as follows: Data normalization and binding operations are performed. Feature parameters extracted and quantified from the aforementioned time-frequency domain analysis (such as center frequency, bandwidth, and impact waveform statistics), load spectrum calculation results, and modal parameters obtained through signal processing methods such as Hilbert-Huang transform (such as damping ratio and mode shape) are organized according to a unified format specification. All these multi-dimensional quantitative feature results are strongly correlated with the specific potential interferometric component pair through a unique coded identifier, ensuring clear data traceability. A relational database structure is designed and initialized as the physical carrier of the quantitative interferometric fingerprint database. A core data table named "Interferometric Mode Feature Table" is created in this database. Each record in this table aims to fully characterize the unique, quantified interferometric mode of a specific potential interferometric component pair. In addition to the primary key field for association and retrieval, this data table also predefines multiple structured data fields according to technical requirements. The feature data is structured and stored in the database. For each potential interference component pair, its bound quantitative feature parameters are filled into the corresponding fields of the data table. For example, the "Unique Identifier of Component Pair" field is filled with an encoding such as "ARM_CYL_01", the "Center Frequency and Bandwidth of Feature Band" field is stored with a numerical pair like "255, 10" (representing a center frequency of 255Hz and a bandwidth of 10Hz), the "Time Domain Statistical Characteristics of Transient Impact Waveform" field is stored with an array containing statistics such as peak value, rise time, and pulse width, and the "Damping Ratio and Mode Vector of Signal Mode" field is stored with the damping ratio value obtained through experimental modal analysis or operational modal analysis and the normalized mode vector coordinate sequence. To achieve the association and fusion of multidimensional data, the initial data table of interference risk, which was previously established for each potential interference component pair and contains information such as static design gaps and dynamic convex hull intersection volume, is linked with the core data table of the quantitative interference fingerprint database created in this step through the key field of "unique identifier of component pair". This forms a complete data knowledge system that integrates static design risk, dynamic kinematic characteristics and measured signal characteristics, providing comprehensive and structured data support for subsequent real-time pattern matching and risk assessment. S4: Based on the matching results of the target interference signal and the quantitative interference fingerprint database, and combined with preset material fatigue characteristic parameters and real-time load spectrum, a fusion risk situation index is generated through calculation, and the currently dominant interference mode type is determined. This step is the core of the method for achieving intelligent diagnosis and comprehensive risk assessment. It does not perform a single judgment, but rather executes a complex calculation process of multi-source information fusion. Specifically, it first identifies the most likely interference mode (i.e., "determines the currently dominant interference mode type") by matching the real-time "target interference signal" with the historically accumulated "quantitative interference fingerprint database". More importantly, it creatively combines the signal matching results characterizing short-term dynamic risk, the "material fatigue characteristic parameters" reflecting medium- and long-term structural damage potential, the "real-time load spectrum", and the risk change trend (differential term) through a specific algorithm, and finally "generates the fusion risk situation index". This index is a scalarized global risk measure that integrates immediate severity and evolution trend, providing a precise and unified basis for the leap from "phenomenon diagnosis" to "risk quantification decision-making". In one specific implementation, the process of generating the fusion risk situation index through fusion calculation is implemented through a multi-level, multi-source information fusion algorithm, as follows: From the real-time matching results of the target interferometric signal and the quantized interferometric fingerprint database, three dimensions of sub-matching degree are extracted: First, the energy proportion of the characteristic frequency band, obtained by calculating the ratio of the energy of the target signal within a specified characteristic frequency band in the fingerprint database to the total energy of the signal; second, the transient impact waveform correlation, obtained by performing cross-correlation analysis on the impact segment of the target signal and the standard impact waveform template stored in the fingerprint database to obtain the maximum cross-correlation coefficient; and third, the signal modal confidence, calculated by comparing the modal shape extracted from the target signal with the modal guarantee criterion value between the standard mode shape vector in the fingerprint database. These three sub-matching degrees are then multiplied by preset contribution weight coefficients and linearly weighted and summed. The preset contribution weight coefficients are determined based on data analysis of a large number of historical interference events, using the analytic hierarchy process (AHP) combined with expert experience. For example, in an implementation targeting a tunnel boring machine (TBM) propulsion system, given that the correlation of the impact waveform is most sensitive to the severity of the collision, it is assigned a weight of 0.5, while the characteristic frequency band energy ratio and signal mode confidence are assigned weights of 0.3 and 0.2, respectively. The initial risk contribution value obtained by weighted summation is finally transformed into a real-time component interference risk value within the range of zero to one using a linear normalization function (such as mapping the value to the interval defined by its historical maximum and minimum values). A standard three-point rainflow counting cycle statistical analysis is performed on the load amplitude sequence in the real-time load spectrum to obtain a load cycle distribution matrix with load amplitude and mean as dimensions. This load cycle distribution matrix is ​​then convolved with the stress-life curve for a specific key structural component (such as the propeller cylinder lug) retrieved from preset material fatigue characteristic parameters. Specifically, a modified Miner linear cumulative damage rule is applied to calculate the damage contribution of each load cycle based on the inverse of the fatigue life corresponding to its amplitude on the stress-life curve, and the summation yields the equivalent fatigue damage degree within the current calculation time window. Introducing a time memory effect, this equivalent fatigue damage degree is added to a historical damage memory factor, defined as the cumulative material damage risk value output in the previous calculation cycle multiplied by a preset historical attenuation coefficient less than 1 (e.g., 0.93, which is set based on material stress relaxation characteristics and engineering experience to reflect the non-immediate recovery of damage). The sum is subjected to a limit processing between zero and one, that is, if it exceeds 1, it is taken as 1, and if it is less than 0, it is taken as 0, thus obtaining the material damage cumulative risk value between zero and one, which represents the fatigue perspective. Multi-dimensional situational fusion calculations are performed. The geometric mean of the obtained real-time component interference risk value and the obtained cumulative material damage risk value is calculated to comprehensively reflect the balanced impact of the two different risk mechanisms. This geometric mean is squared to amplify the numerical differences in high-risk states, yielding the basic situational component. Simultaneously, the sequence of the fusion risk situation index over the most recent time windows is numerically differentiated, and its rate of change is calculated as a differential term reflecting the risk change trend. The absolute value of this differential term is then cubed to smooth out sharp fluctuations in the trend while preserving its directional information, yielding the trend change component. The basic situational component and the trend change component are added to obtain an unweighted preliminary situational value. This preliminary situational value is multiplied by an adaptive weighting coefficient dynamically adjusted based on historical control effect data. This adaptive weighting coefficient is iteratively updated according to the actual reduction effect of the risk index after control actions under similar past operating conditions (e.g., increasing the weight to enhance early warning sensitivity if the effect is significant, and decreasing the weight to avoid over-response if the effect is poor), thus obtaining the final, dynamically adaptive fusion risk situation index.

[0024] Integration and Output. The above calculation process is integrated into an automated data flow. Upon receiving data from a new analysis time window, the aforementioned steps are executed sequentially, outputting a unified, quantitative fusion risk situation index. For example, in a scenario where interference is detected between the anchor drill arm and the cylinder support, this index may comprehensively reflect the intensity of the high-frequency interference signal (real-time risk value 0.75), the cumulative damage state of the lug material under current and historical loads (cumulative risk value 0.60), and the rapidly increasing risk trend (positive differential term). The final fusion calculation yields a higher situation index (e.g., 0.82), providing precise quantitative basis for subsequent control decisions.

[0025] In one specific implementation, the multi-dimensional situational fusion calculation is performed according to the following specific steps to generate a fused risk situational index. First, a comprehensive risk benchmark is calculated by taking the real-time risk value of component interference and the cumulative risk value of material damage obtained in the current calculation cycle, and calculating their geometric mean. For example, if the real-time risk value of component interference is 0.75 and the cumulative risk value of material damage is 0.60, the geometric mean is the square root of the product of these two values, which is approximately 0.6708. This operation aims to balance the combined impact of risks from two different physical sources, avoiding extreme dominance by any one risk value. Second, basic situational components are constructed by squaring the geometric mean obtained in the previous step. Taking the above value as an example, the square of 0.6708 is approximately 0.45, yielding the basic situational components. The mathematical effect of squaring is to nonlinearly amplify high geometric means, making high-risk comprehensive states more significantly reflected in subsequent calculations, thereby improving the sensitivity of the situational index to dangerous conditions. The risk change trend is extracted and processed by performing central difference on the fusion risk situation index sequence generated in the most recent five calculation periods to calculate the differential term of the risk change trend in the current period. The absolute value of this differential term is then cubed; for example, if the absolute value of the differential term is 0.125, its cube root is 0.5, yielding the trend change component. The cube root operation effectively smooths out drastic changes in trend values ​​caused by short-term fluctuations, while preserving the directionality of the trend (by retaining the original sign of the differential term in subsequent addition steps), making the contribution of the trend component more robust. A preliminary situation value is synthesized by algebraically adding the previously calculated basic situation component and the trend change component; for example, adding the basic situation component 0.45 to the trend change component 0.5 yields an unweighted preliminary situation value of 0.95. Finally, a final index is generated using dynamic weights by multiplying the initial situation value by an adaptive weight coefficient. This adaptive weight coefficient is dynamically updated based on a continuous assessment of the historical control effects of the potential intervention component. In one example, if historical data shows that control commands executed under three similar risk situations in the last three periods have effectively reduced the risk index by more than 30%, the adaptive weight coefficient may be increased to 1.1 to enhance early warning; conversely, if the control effect is poor, the coefficient may be decreased to 0.9 to smooth index fluctuations. The initial situation value of 0.95 is multiplied by, for example, the current weight coefficient of 1.0 to obtain the final fused risk situation index of 0.95, completing this round of multi-dimensional situation fusion calculation. In one specific implementation, determining the currently dominant interference mode type is accomplished through a refined pattern recognition and decision-making process, specifically as follows: Multidimensional feature similarity is calculated. For each identified potential interference component pair, the standardized feature parameter set stored in the quantized interference fingerprint database is retrieved, including the center frequency and bandwidth range of the characteristic frequency band, the standard template of the transient impact waveform, and the mode shape vector of the signal mode. Simultaneously, the corresponding actual feature parameters are extracted from the real-time acquired target interference signal. Similarity is quantified by calculating the closeness between the two in multiple dimensions: in the frequency domain, the energy concentration of the actual signal within the standard characteristic frequency band and the Bhattacharyya coefficient of the standard energy distribution in that band are calculated; in the time domain, the cross-correlation coefficient between the actual impact segment and the standard waveform template is calculated; in the modal dimension, the modal guarantee criterion value between the extracted mode shape vector and the standard mode shape vector is calculated. The closeness calculation results of these three dimensions are geometrically averaged to obtain the multidimensional feature similarity of the component pair in this event. The comprehensive matching degree is calculated by coupling energy weights, introducing a weighting factor that reflects the proportion of signal energy of the component pair in the target interferometric signal. This weighting factor is calculated based on the energy contribution principle in signal processing. By filtering and reconstructing the target interferometric signal using standard characteristic frequency bands from a quantized interferometric fingerprint database, the proportion of signal energy belonging to the currently evaluated component pair in the total signal energy is estimated. For example, if analyzing the component pair of the anchor drill arm and the propulsion cylinder support, whose standard characteristic frequency band is 240-260Hz, the ratio of the filtered signal energy of the target interferometric signal within the 240-260Hz band to the total signal energy across the entire frequency band is calculated and used as its weighting factor. The obtained multidimensional feature similarity is then productted and coupled with this weighting factor to obtain a comprehensive matching degree. This operation aims to assign higher matching confidence to interferometric modes with large signal energy contributions. The confidence threshold for the interference pattern is set and applied based on statistical analysis. The threshold is determined by statistical analysis of historical normal operating conditions and confirmed interference events, combined with the analysis of receiver operation characteristic curves in pattern recognition. Specifically, a large number of historical data samples are collected, and the comprehensive matching degree of each sample on different potential interference component pairs is calculated. By analyzing the matching degree distribution between confirmed interference samples and background noise samples, the inflection point value that achieves the best balance between the true positive rate and the false positive rate is selected as the threshold. In an exemplary implementation, after analyzing more than 1000 hours of tunneling data, this threshold is set to 0.70. In real-time judgment, the comprehensive matching degree calculated for all current potential interference component pairs is compared with this preset threshold of 0.70. The execution decision-making logic determines the dominant mode. Based on the comparison results, a preset logical judgment is executed: If the overall matching degree of all potential interfering component pairs does not exceed 0.70, it is determined that there is currently no dominant interference mode, which may be background noise or an unknown identification event; if there is a single overall matching degree exceeding 0.70, for example, only the "anchor drill arm - propulsion cylinder support" pair has an overall matching degree of 0.85, then the interference mode of the component pair corresponding to this value is directly determined as the current dominant interference mode type; if there are multiple overall matching degrees exceeding 0.70, for example, the "anchor drill arm - propulsion cylinder support" pair is 0.85, and the "segment assembler - shield tail sealing shell" pair is 0.78, then following the conservative early warning principle, the interference mode of the component pair corresponding to the highest value of 0.85, i.e., the interference mode of the anchor drill arm and the propulsion cylinder support, is determined as the current dominant interference mode type, and its identifier is output to subsequent control decision steps. S5: Based on the integrated risk situation index and interference mode type, the system queries the preset adaptive control strategy matrix to dynamically determine and execute tiered control instructions and their control parameters. This step serves as the decision-making and execution hub for transforming risk assessment conclusions into specific control actions. It introduces a core intelligent decision-making component—the "preset adaptive control strategy matrix." This matrix uses the "integrated risk situation index" (risk magnitude) and "interference mode type" (risk nature) as two-dimensional input indices. By "querying" this matrix, the system can dynamically determine the optimal or suboptimal response combination, i.e., "tiered control instructions and their control parameters," from a preset strategy library based on the current specific risk state (e.g., high-frequency minor scratch vs. low-frequency severe collision). These instructions may include one or more measures such as applying anti-phase acoustic vibration, adjusting component stiffness, and modifying tunneling waveforms, and their intensity, timing, and other parameters are also dynamically calculated. Subsequently, the instructions are issued and executed, achieving proactive, precise, and dynamic suppression of "ghost interference." In one specific implementation, the process of dynamically determining and executing graded control commands and their control parameters in step S5 is achieved through an automated process that integrates strategy query, parameter calculation, and command coordination, as follows: Intelligent query and command retrieval of the strategy matrix are performed. Based on the real-time calculated fusion risk situation index (e.g., 0.82) and the determined current dominant interference mode type (e.g., "anchor drill arm - propulsion cylinder support"), a location query is performed in a pre-built and continuously optimized adaptive control strategy matrix. This adaptive control strategy matrix is ​​a two-dimensional decision table with risk index ranges as rows and different interference mode types as columns. Each cell, i.e., the strategy unit, stores an ordered set of one or more control command identifiers recommended for that specific risk-mode combination. For example, querying the strategy unit where the risk index range [0.8, 0.9) intersects with the "anchor drill arm - propulsion cylinder support" mode may retrieve two control command identifiers to be executed: "anti-phase acoustic vibration control" and "propulsion waveform adjustment". The system performs dynamic real-time calculation of control parameters. Based on the specific preset risk level range of the integrated risk situation index (e.g., 0.82 belongs to the "high risk" range), it calls the parameter calculation rule library associated with the strategy unit. This rule library defines the parameter calculation formula or mapping table corresponding to each control command identifier, and its input is the current accurate risk situation index value. For example, for the "anti-phase acoustic vibration control" command, its output parameter (such as the actuator force amplitude) is dynamically calculated according to a linear function proportional to the risk index value. For the "propulsion waveform adjustment" command, its output parameter (such as the thrust reduction percentage within a specific cutterhead angle range) is directly obtained according to a preset parameter mapping table corresponding to the risk level range. This generates a set of specific control parameter values ​​that precisely correspond to each control command identifier. The coordinated issuance and execution sequence control of execution commands, according to preset execution logic (e.g., for high-frequency impact interference, priority is given to issuing anti-phase acoustic vibration control commands to quickly suppress vibration, followed by issuing propulsion waveform adjustment commands for root load adjustment), transmits the control command identifiers and their corresponding specific control parameter values ​​synchronously or sequentially to the underlying programmable logic controllers or dedicated motion controllers corresponding to the tunnel boring machine and support equipment via industrial fieldbus or real-time Ethernet protocol. This drives the physical actuators to complete closed-loop control actions. After receiving the commands and parameters, the underlying controller drives the corresponding physical actuators to perform precise actions; for example, the anti-phase acoustic vibration control command drives the piezoelectric actuators installed on the anchor drill arm to generate specific amplitude and phase canceling vibrations; the propulsion waveform adjustment command modifies the control signal of the hydraulic propulsion system through the tunnel boring machine's main control system, fine-tuning the thrust at the millisecond level in a specific phase; after all actions are executed, their state feedback signals are re-acquired, forming a complete closed loop from decision-making to execution to verification, providing a data basis for the subsequent strategy optimization in step S6. In a specific example, when the fusion risk situation index is 0.82, it is classified as a "high-risk" interval according to preset rules, and the current dominant interference mode type is a high-frequency collision of "anchor drill arm - propulsion cylinder support," the system calls the parameter calculation rule library associated with this strategy unit for dynamic calculation. This rule library defines the parameter determination logic for each control command under different risk levels: For the "propulsion waveform adjustment" command, in the "high-risk" interval, its parameter is to reduce the hydraulic propulsion system's thrust reference value by one percentage point within the angle intervals corresponding to the phase of interference occurrence: 90° to 135° and 270° to 315°. This percentage is calculated by linear interpolation in a pre-mapped "high-risk" interval parameter table (e.g., an index value of 0.80 corresponds to a 15% reduction, and 0.90 corresponds to a 25% reduction). This calculation results in a 20% thrust reduction. Simultaneously, for the "anti-phase acoustic vibration control" command, its force amplitude parameter is determined by the formula... Dynamic calculation is performed, where K is the preset gain coefficient for the interference mode and R is the current fusion risk situation index. The calculated force amplitude is a specific value, thereby generating a set of precise control parameter values ​​that match the severity of high risk in real time.

[0026] S6: Using the execution effect data of the hierarchical control commands as feedback, the parameters of the corresponding decision logic units in the adaptive control strategy matrix are updated. Based on the updated model, the support control timing and maintenance window for subsequent tunneling cycles are dynamically optimized and scheduled. This step endows the entire system with the ability to learn and continuously optimize, forming a complete intelligent closed loop of "perception-decision-execution-learning". It does not end with the completion of a single control, but uses the "execution effect data" of each control (i.e., changes in the risk index after control, equipment status response, etc.) as valuable feedback information. Using this feedback, the system can reversely "update the parameters of the corresponding decision logic units in the adaptive control strategy matrix", which means that the decision strategy will continuously self-correct and optimize with the accumulation of engineering practice data, becoming more and more accurate and efficient. At the same time, the results of this learning and optimization will be further applied to the forward-looking planning of future work, namely "dynamically optimizing and scheduling the support control timing and maintenance window for subsequent tunneling cycles", thereby realizing the evolution from passive response to active prevention, from local optimization to global collaboration, greatly improving the adaptability and long-term effectiveness of the entire support measure design method. In one specific implementation, the execution process of step S6 is achieved through a closed-loop learning process based on data-driven and iterative optimization, as follows: Data collection and quantitative evaluation of execution effect: After the graded control command issued in step S5 is executed, multi-source mixed strain wave signals are continuously collected within several subsequent analysis time windows, and the fusion risk situation index is recalculated. By comparing the attenuation magnitude of this index before and after the control action, the time required for stabilization, and the degree of peak reduction in the real-time load spectrum of key structural components, the immediate effectiveness score of this control action is comprehensively calculated. For example, for a counter-phase acoustic vibration control triggered to cope with the interference of the "anchor drill arm-propulsion cylinder support", calculations show that after control, the energy of the component to the relevant target interference signal decreased by 65%, and the fusion risk situation index decreased from 0.82 to 0.45 within 3 seconds. Based on this, the immediate effectiveness score of this control is calculated to be 0.85 (out of 1.0). Based on feedback data, the adaptive control strategy matrix is ​​iteratively updated online. Using the interval of the fusion risk situation index and the type of intervention mode used when triggering the control decision as an index, the corresponding decision logic unit in the adaptive control strategy matrix is ​​located. The calculated real-time performance score is compared with the historical average performance score recorded in the decision logic unit to calculate the performance deviation. Based on the sign and magnitude of the performance deviation, combined with a preset learning rate parameter, the priority of the control instructions stored in the decision logic unit or the associated parameter calculation rules are fine-tuned. For example, if the effect of the current anti-phase acoustic vibration control is much better than the historical average, the priority weight of the instruction under this risk-mode combination is increased, and the gain coefficient in its parameter calculation formula is slightly increased, so that it is used more preferentially in similar situations and the control intensity is moderately enhanced. The parameters of the driving correlation model are synchronously and adaptively updated, feeding back the entire chain of data on the control process and its effects, including the original strain wave signal, matching results, calculated risk value, executed instructions and parameters, and final performance, to the component kinematic model, the quantitative interferometric fingerprint database, and the damage prediction model. For example, by using the signal characteristics when actual interference occurs, the characteristic frequency band descriptions of the corresponding component pairs in the quantitative interferometric fingerprint database are updated with confidence weights, making the fingerprint database features closer to the real working conditions. At the same time, based on the actual correspondence between load and damage, the parameters in the damage prediction model associated with the component pair are corrected to improve its prediction accuracy.The system dynamically optimizes the scheduling of subsequent tasks, conducting forward-looking analyses of upcoming tunneling cycles based on the updated adaptive control strategy matrix and various prediction models. It predicts potential interference-prone conditions (such as uneven strata) based on geological forecast data and the tunneling plan, and generates recommended control plans for these future conditions based on the updated strategies and models. Simultaneously, it dynamically optimizes the timing of subsequent support control (e.g., proactive vibration suppression at specific tunneling rings) and planned maintenance windows (e.g., inspecting and tightening specific cylinder lugs before the expected accumulated damage reaches a threshold), thereby achieving advanced decision support from passive response to proactive prediction and planning. In a specific example, suppose the historical average performance score of the decision logic unit for the "anchor drill arm - propulsion cylinder support" interference mode in the high-risk range is 0.70. After executing the combined command of anti-phase acoustic vibration control and propulsion waveform adjustment, the calculated immediate performance score is 0.85, and the performance deviation is +0.15. The preset learning rate parameter is 0.1. According to the update rule... The priority of the control commands stored in this unit is fine-tuned: the priority weight of the anti-phase acoustic vibration control command is updated from 0.7 to 0.715; at the same time, the associated parameter calculation rules (force amplitude formula) are also adjusted. The gain coefficient K in the module was adjusted synchronously, with the K value updated from 1200 to 1218. The historical average performance score of this unit was also updated. .

[0027] The above formulas are all dimensionless calculations, and the preset parameters in the formulas should be set by those skilled in the art according to the actual situation.

[0028] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A design method for support measures in a small-diameter TBM tunnel, characterized in that, Includes the following steps: S1: Distributed strain wave sensors are installed on the base of key structural components and support equipment of the tunnel boring machine to collect multi-source mixed strain wave signals and obtain the pre-set material fatigue characteristic parameters of the key structural components. S2: Based on a predefined component kinematic model, the target interference signal associated with the potential interference component pair is separated from the multi-source mixed strain wave signal; S3: Perform time-frequency domain analysis on the target interference signal, extract and quantize its characteristic parameters to generate the real-time load spectrum of key structural components, and establish a quantitative interference fingerprint library for each potential interference component based on the quantization results; S4: Based on the matching results of the target interferometric signal and the quantitative interferometric fingerprint database, and combined with the preset material fatigue characteristic parameters and real-time load spectrum, the fusion risk situation index is generated by fusion calculation, and the current dominant interferometric mode type is determined. S5: Based on the integrated risk situation index and the type of intervention mode, query the preset adaptive control strategy matrix, dynamically determine and execute the hierarchical control instructions and their control parameters; S6: Use the execution effect data of the hierarchical control command as feedback to update the parameters of the corresponding decision logic unit in the adaptive control strategy matrix, and dynamically optimize the timing of support control and maintenance window of subsequent tunneling cycles based on the updated model.

2. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, Step S1, which involves acquiring multi-source hybrid strain wave signals, specifically includes: Fiber grating sensors or piezoelectric sensors are used as distributed strain wave sensors; Each distributed strain wave sensor is rigidly connected and placed in the stress concentration area of ​​key structural components and support equipment base. Synchronous data acquisition is performed on all distributed strain wave sensors at a sampling frequency of not less than 10kHz; Each distributed strain wave sensor acquires a strain wave signal and assigns a corresponding spatial location code to form a multi-source hybrid strain wave signal.

3. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, Step S2, separating the target interference signal, specifically includes: The expected relative kinematic characteristics of each potential interfering component pair are calculated based on the component kinematic model; The blind source separation algorithm was used to process the multi-source mixed strain wave signal, and several independent source signals were initially separated. The time-frequency characteristics of independent source signals are matched with the expected relative kinematic characteristics. Independent source signals with a matching degree exceeding the preset interference feature matching degree threshold are identified and extracted as target interference signals associated with specific potential interference component pairs.

4. The design method for support measures of a small-diameter TBM tunnel according to claim 3, characterized in that, Potential interference pairings are identified in the following way: Based on the digital prototype model of the tunnel boring machine and support equipment, identify all component combinations with relative motion; For each component assembly, a parameterized multibody dynamics numerical simulation model is established based on the kinematic constraints determined by its actual mechanical connection method. Using a parameterized multibody dynamics numerical simulation model, the convex hull boundary of the spatial point cloud set formed by the motion trajectories of all feature points on the surface of each moving part during its entire working cycle is calculated. The component pairs that have static design gaps between components obtained from the analysis of the digital prototype model that are less than the safety threshold and whose calculated convex hull boundaries intersect are predefined as potential interference component pairs, and their initial interference risk profiles are established.

5. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, Step S3, establishing the quantitative interferometric fingerprint database, specifically includes: The quantization results of the feature parameters are bound to the corresponding potential interferometer pairs; A relational database is established as a quantitative interference fingerprint library, in which each record represents a unique interference pattern of a potential pair of interfering components; Each record contains the following data fields: unique identifier for the component, center frequency and bandwidth of the characteristic frequency band, time-domain statistical characteristics of the transient impact waveform, and damping ratio and mode shape vector of the signal mode. The initial profile of the interference risk of each potential interference component pair is associated with its record in the quantitative interference fingerprint database.

6. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, The specific process of generating the fusion risk situation index in step S4 includes: Three sub-matching degrees are extracted: characteristic frequency band energy ratio, transient impact waveform correlation, and signal mode confidence. The three sub-matching degrees are multiplied by preset contribution weight coefficients and then linearly weighted and summed to obtain the initial risk contribution value. The initial risk contribution value is then mapped to the zero-to-one interval through a normalization function to obtain the real-time risk value of component interference. Rainflow counting statistics are performed on the load amplitude sequence in the real-time load spectrum to obtain the load cycle distribution; the load cycle distribution is convolved with the stress-life curve in the preset material fatigue characteristic parameters to calculate the equivalent fatigue damage degree under the current load cycle; the equivalent fatigue damage degree is added to a historical damage memory factor, and the sum is then limited to the range of zero to one to obtain the material damage cumulative risk value; where the historical damage memory factor is equal to the material damage cumulative risk value of the previous calculation cycle multiplied by a preset historical attenuation coefficient less than one; The real-time risk value of component interference, the cumulative risk value of material damage, and a differential term reflecting the trend of risk change are fused together in a multi-dimensional situational analysis to obtain the final fused risk situational index.

7. The design method for support measures of a small-diameter TBM tunnel according to claim 6, characterized in that, Multi-dimensional situational fusion calculation is performed according to the following steps: Step 1: Calculate the geometric mean of the real-time risk value of component interference and the cumulative risk value of material damage; Step 2: Squaring the geometric mean yields the basic situational components; Step 3: Calculate the cube root of the absolute value of the differential term of the risk change trend to obtain the trend change component. Step 4: Add the basic situational component to the trend change component to obtain the unweighted preliminary situational value; Step 5: Multiply the preliminary situation value by the adaptive weighting coefficient to obtain the final fused risk situation index.

8. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, Step S4, determining the currently dominant interferometry mode type, specifically includes: For each potential pair of interfering components, calculate the multidimensional feature similarity between the feature parameters stored in its quantized interferometric fingerprint database and the corresponding feature parameters of the target interferometric signal; By coupling the multidimensional feature similarity with a weighting factor that reflects the proportion of energy of the signal components in the target interference signal, a comprehensive matching degree is obtained. Set an interference mode confidence threshold and compare the overall matching degree of all potential interference component pairs with this confidence threshold; If there is a unique comprehensive matching degree that exceeds the confidence threshold, then the interference mode of the potential interfering component pair corresponding to the comprehensive matching degree is determined as the currently dominant interference mode type; If there are multiple combined matching degrees that exceed the confidence threshold, the interference mode of the potential interfering component pair corresponding to the one with the highest value will be determined as the currently dominant interference mode type.

9. The design method for support measures of a small-diameter TBM tunnel according to claim 1, characterized in that, Step S5, which involves dynamically determining and executing tiered control commands and their control parameters, specifically includes: Query the preset adaptive control strategy matrix to find the strategy unit corresponding to the current fusion risk situation index and intervention mode type, and read one or more control instruction identifiers to be executed stored therein; Based on the preset risk level range of the integrated risk situation index, a set of specific control parameter values ​​corresponding to each control instruction identifier are dynamically calculated from the strategy unit. According to the preset execution logic, the control command identifier and its corresponding specific control parameter value are synchronously or sequentially sent to the bottom controller of the tunnel boring machine and support equipment to drive the corresponding physical actuators to complete the control action.

Images