A method and system for monitoring wear of a shield cutter

Abstract

The application discloses a kind of wear monitoring method and system of shield cutter, belong to cutter quality monitoring technical field, comprising: to scanning frequency is differentially encoded, and to cutter is zoned scanning, obtain the vibration signal of target cutter, and carry out multi-scale feature extraction, generate time-frequency energy spectrum, avoid single scale feature to the misjudgment of complex wear mode;Acquire cutter temperature field distribution, identify local over-temperature area, associated to corresponding cutter coordinate, and carry out wear interactive analysis, calculate interactive compensation matrix, to generate the wear characteristic of target cutter;Build wear mapping library, carry out wear state identification, and dynamically adjust classification threshold, once the characteristic value of target cutter continuously exceeds classification threshold, marked as abnormal wear, trigger cutter changing mechanism, generate cutter changing sequence, inhibit wear chain reaction.

Description

Technical Field

[0001] This invention relates to a method and system for monitoring the wear of tunnel boring machine (TBM) cutters, belonging to the field of cutter quality monitoring technology. Background Technology

[0002] As the core equipment for tunnel construction, the wear condition of the tunnel boring machine's cutters directly affects construction efficiency and operational safety. Traditional cutter wear monitoring mainly relies on manual inspection or indirect inference based on tunneling parameters. This method suffers from low detection accuracy, poor real-time performance, and insufficient adaptability to complex working conditions. In particular, under extreme conditions such as deep-sea high pressure and uneven strata, frequent opening of the tunnel poses a very high risk. Furthermore, the indirect inference method based on parameters is easily affected by geological changes and equipment conditions, making it difficult to meet the urgent need for real-time and accurate monitoring in engineering projects.

[0003] During the cutting process, the interaction between the cutting tool and the rock and soil mass generates vibration signals. These signals contain rich information about the tool wear characteristics and are characterized by strong real-time performance and large information content. They can directly reflect the dynamic working condition of the tool. By extracting the time-frequency domain features of the vibration signals and combining them with machine learning algorithms, high-precision identification of the tool wear state is achieved.

[0004] However, existing technologies do not consider the complex wear interaction effects caused by the collaborative operation of multiple tools on the tool barrel. Specifically, when multiple tools in the tool barrel are cutting synchronously, if one or more tools in a local area experience abnormal wear, the stress state and cutting behavior of adjacent tools will change significantly. This will cause the nonlinear vibration distortion generated to be transmitted to the dedicated sensor through the tool barrel structure, resulting in the contamination of the original vibration characteristics of healthy tools. Meanwhile, the signal characteristics of the actual worn tools will still be masked or distorted by the interaction effect between adjacent tools. This "load transfer-dynamic response" coupling effect caused by worn tools seriously interferes with the traditional monitoring model based on the assumption of independent single tool, ultimately leading to wear location failure, misjudgment of degree, and missed early warning, which greatly restricts the ability to make accurate tool replacement decisions and control construction risks. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the present invention aims to provide a method and system for monitoring the wear of tunnel boring machine (TBM) cutters. By acquiring the vibration signal of the target cutter, dynamic correction is performed to ensure scanning accuracy. An interactive compensation matrix is ​​used to quantify the influence of adjacent cutters. Topological network is used for directional decoupling to restore the characteristics of a single cutter. Combined with working condition adaptation, recognition is improved. Cutter replacement is optimized to suppress chain reactions, and signal aliasing and interactive interference problems are solved.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] A method for monitoring the wear of tunnel boring machine cutters, comprising:

[0008] A laser vibration measuring device is installed behind the tool inside the tool barrel to scan the target tool, obtain the vibration signal of the target tool, and perform multi-scale feature extraction to generate a time-frequency energy spectrum.

[0009] The temperature field distribution of the tool barrel is obtained, local overheating areas are identified, and associated with the corresponding tool coordinates. Wear interaction analysis is performed, and the interaction compensation matrix is ​​calculated to generate the wear characteristics of the target tool.

[0010] A wear mapping library is constructed to identify wear conditions and dynamically adjust the classification threshold. Once the feature value of the target tool continuously exceeds the classification threshold, it is marked as abnormal wear, triggering the tool change mechanism and generating a tool change sequence.

[0011] Specifically, the steps of multi-scale feature extraction include:

[0012] Obtain the physical parameters of each tool in the tool barrel and generate a tool type library;

[0013] Based on historical wear data, wear risk is classified according to tool type, and corresponding scanning priority and scanning interval are set to generate a scanning list;

[0014] The laser vibration measurement device is aligned with the target tool, and the time of each device is calibrated to generate a single-tool scanning command. The reflection signal of the target tool is collected, noise is filtered, and a preliminary filtered signal is obtained.

[0015] A noise canceller is constructed using the least mean square algorithm to obtain the vibration signal of the target tool.

[0016] Specifically, the steps of multi-scale feature extraction also include:

[0017] Calculate the target variance of the vibration signal, obtain the wear threshold, if the target variance is less than the wear threshold, it is a stable stage; otherwise, it is a rapid stage, and generate a wear stage identifier.

[0018] Based on the stage parameter library, the segment duration of the corresponding stage is called, the segment signal is extracted, and wavelet threshold denoising is combined to generate the vibration segment signal of the target tool.

[0019] The STFT parameter library is called to generate a low-frequency matrix, and the wavelet parameter library is called to configure the corresponding decomposition parameters, thereby generating a high-frequency matrix.

[0020] Calculate the low-frequency difference and high-frequency difference to obtain the global difference, thereby generating the low-frequency contribution and high-frequency contribution.

[0021] The weight correction library is called to extract the correction coefficients, calculate the low-frequency weights and high-frequency weights, and perform weighted calculations to generate a time-frequency energy spectrum.

[0022] Specifically, the steps for identifying localized overheating areas include:

[0023] The temperature field image of the cutter barrel is acquired in real time and preprocessed.

[0024] The Sobel operator is used to calculate the pixel gradient. Local maxima along the gradient direction are preserved by non-maximum suppression. Double thresholds are set to distinguish strong edges, weak edges and non-edges, and an edge image is generated.

[0025] Perform a Hough transform on the edge image to generate a set of pixels for the blade edge region;

[0026] Calculate the temperature gradient of each pixel, and take the maximum temperature gradient value in the cutting edge area as the local temperature gradient of the cutting edge of the target tool.

[0027] Obtain the upper limit of the normal temperature gradient, set the dynamic coefficient, and calculate the over-temperature threshold by combining it with the vibration amplitude of the target tool;

[0028] If the local temperature gradient of the cutting edge is greater than the over-temperature threshold and the duration exceeds a preset time interval, it is determined to be a local over-temperature region, and actual spatial coordinates are generated to construct an over-temperature list.

[0029] Specifically, the steps of wear interaction parsing include:

[0030] Construct a multidimensional feature set of the target tool, calculate the Euclidean distance between the target tool and other tools in the tool barrel, combine the neighborhood threshold to generate a target neighborhood set, and construct neighbor tool interaction features;

[0031] Using the actual wear amount of the target tool as the dependent variable and the multidimensional feature set and the neighbor tool interaction features as independent variables, an association model is constructed to generate the predicted wear amount of the target tool.

[0032] Set a two-level wear threshold to classify the wear state of the target tool, including normal, light, and heavy.

[0033] Based on any neighboring tool of the target tool, obtain the interaction coefficient with the neighboring tool in the association model, and perform normalization processing to obtain the weight of each feature in the neighboring tool interaction features;

[0034] Each feature in the neighboring tool interaction features is weighted to obtain the compensation factor of the neighboring tool, and combined with the target neighborhood set to generate the interaction compensation matrix of the target tool.

[0035] Specifically, the steps of wear interaction parsing also include:

[0036] Using the cutting tools on the tool barrel as network nodes, each node is assigned node attributes, including wear intensity and node degree. Adjacent tool pairs are used as edges, and the corresponding compensation factors are used as edge weights to construct the interactive topology network of the target tool.

[0037] By combining compensation factors, the PageRank algorithm is used to calculate node influence, and key interference sources are screened out by combining influence thresholds, and key interference paths are marked.

[0038] For the target tool, only the adjacent tools on the critical interference path are retained, the interference component of a single adjacent tool is calculated, and the components are accumulated based on the adjacent tools to obtain the quantitative interference amount of the target tool.

[0039] Remove the directional interference from the time-frequency energy map of the target tool to obtain the actual time-frequency characteristics, and calculate the feature fidelity. Once the feature fidelity is not less than the decoupling threshold, recalculate the path weight.

[0040] For the decoupled image, features are extracted to construct the wear characteristics of the target tool.

[0041] Specifically, the steps for wear condition identification include:

[0042] The historical wear characteristics of each tool in the tool barrel are obtained, and cluster analysis is performed on the wear characteristics of the same type and state to obtain the baseline characteristics, thereby generating a wear mapping library;

[0043] Based on the wear characteristics of the target tool, reference features are invoked, and the similarity between the wear characteristics and the reference features is calculated. The wear state corresponding to the maximum similarity is selected as the anchoring state.

[0044] Real-time operating parameters are acquired, operating condition correction factors are calculated, and the benchmark features of the anchoring state are corrected to form an operating condition adaptation benchmark. The similarity between the damage features and the operating condition adaptation benchmark is calculated.

[0045] If the similarity is improved and the working condition adaptation benchmark is still within the normal fluctuation range of the anchoring state, the original anchoring state is maintained; otherwise, the benchmark feature matching and screening are performed again.

[0046] Specifically, the steps for generating the tool change sequence include:

[0047] The worn tool and its corresponding classification threshold are obtained. Based on the compensation factor of the worn tool, the interference weight of the adjacent tool is calculated, and the classification threshold is corrected to obtain the anomaly judgment threshold.

[0048] The real-time characteristic value of the worn tool is obtained. If the real-time characteristic value is less than the anomaly judgment threshold, it is judged as a slight anomaly; otherwise, it is judged as a severe anomaly.

[0049] For mild abnormalities, if continuous If the real-time feature value within a window exceeds the anomaly detection threshold, it is judged as primary abnormal wear. For severe anomalies, if the value is continuously... If the real-time feature value within a window exceeds the anomaly detection threshold, it is determined to be an emergency abnormal wear. Simultaneously, tools with severe wear are marked as near-abnormal wear, ultimately generating a list of abnormal wear tools. The number of decision windows required to determine primary abnormal wear. To determine the number of decision windows required for emergency abnormal wear, , All settings are based on statistical verification of historical data, and In the embodiments, take , .

[0050] Prioritize abnormally worn tools, verify the position of candidate tool changers, and generate tool change sequences using a greedy algorithm;

[0051] During tool change, the original load percentage of the tool change area is calculated, and combined with the reserved safety redundancy load, the load redundancy of the non-tool change area is calculated to generate regional adjustment commands.

[0052] A wear monitoring system for tunnel boring machine (TBM) cutters includes: a separation module, a monitoring module, a wear analysis module, and a cutter replacement identification module;

[0053] The separation module is used to scan the target tool and obtain vibration signals by setting a laser vibration measuring device behind the tool inside the tool barrel;

[0054] The monitoring module is used to construct a time-frequency energy spectrum and collect the temperature field of the cutter barrel in real time to identify local overheating areas;

[0055] The wear analysis module is used to construct wear mapping relationships, generate interactive compensation matrices, construct a topology network with the tool as nodes, and correct the time-frequency energy spectrum.

[0056] The tool change identification module is used to initially determine the wear status, dynamically adjust the classification threshold in combination with the interactive compensation matrix, identify abnormally worn tools, and generate a batch tool change sequence.

[0057] The beneficial effects of this invention are:

[0058] By scanning the target tool with a laser vibration meter and combining time-division multiplexing and narrowband filtering, the coupling and aliasing of signals from multiple tools are avoided at the source, achieving precise separation of single-tool vibration signals and solving the problem of single-tool features being submerged. Low-frequency trends and high-frequency impact features are extracted through multi-scale time-frequency transformation, and temperature field distribution is obtained by combining infrared thermal imaging to construct a multi-dimensional feature system, accurately locate the wear area, build an association model and interactive compensation matrix, quantify the load transfer impact between adjacent tools, and identify key interference paths through topological network to achieve weighted correction of time-frequency energy spectrum, eliminate interference from adjacent tools, restore the true wear characteristics of single tools, and avoid the problems of healthy tool signals being contaminated and worn tool features being masked. The tool replacement sequence is optimized based on dynamic threshold and load balance, combined with working condition adaptive correction, to ensure accurate identification of wear status and timely tool replacement decision-making, effectively suppress wear chain reaction, and improve the wear monitoring accuracy and construction risk control capability in complex tool barrel environments. Attached Figure Description

[0059] Figure 1 This is a schematic diagram of a method for monitoring the wear of tunnel boring machine cutters.

[0060] Figure 2 This is a flowchart of the multi-scale feature extraction process in this invention;

[0061] Figure 3 This is a flowchart illustrating the identification of local overheating regions in this invention;

[0062] Figure 4 This is a flowchart of the wear interaction analysis in this invention. Detailed Implementation

[0063] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0064] Example 1

[0065] like Figure 1 The diagram shows a flowchart of a method for correcting the wear coefficient of a tunnel boring machine cutter according to an embodiment of the present invention. The method includes steps S101 to S114, wherein:

[0066] S101, Obtain vibration signal: An independent laser vibration measurement device is set behind each tool inside the tool barrel to directly scan the target tool, superimpose narrowband optical filtering to suppress dust scattering noise, and combine lock-in amplification technology to extract vibration signal, so as to realize non-contact global vibration monitoring to obtain the vibration signal of the target tool.

[0067] S102, Multi-scale feature extraction: Low-frequency trend features are extracted by STFT and high-frequency impact features are extracted by wavelet packet decomposition to generate time-frequency energy spectrum, avoiding the omission of complex wear patterns by single-scale features.

[0068] S103, Obtain temperature field distribution: Based on the infrared thermal imaging arranged on the cutter barrel, the temperature field distribution of the cutter barrel is obtained simultaneously;

[0069] S104, Identify local over-temperature areas: Use Canny operator edge detection combined with Hough transform to identify the tool cutting edge area, extract the local temperature gradient of each tool, identify local over-temperature areas, and associate them with the corresponding tool coordinates;

[0070] S105, Wear Interaction Analysis: Based on time-frequency energy spectrum and temperature gradient data, combined with historical wear records, a mapping relationship including wear, temperature and vibration is established, and wear interaction analysis is performed;

[0071] S106, Interactive Compensation Matrix: The compensation factor between any two tools is calculated through multiple regression analysis to generate an interactive compensation matrix, which quantifies the impact of neighborhood wear on the current tool. At the same time, a tool topology graph is constructed with the tool as a node.

[0072] S107, Generate Wear Features: The interaction compensation matrix is ​​used as the weight of the edge to perform weighted correction on the time-frequency energy spectrum, eliminate the interference of adjacent tool wear on the current tool vibration signal, restore the true wear features of a single tool, and thus generate the wear features of the target tool, solving the problem that the wear features of a single tool are masked due to signal distortion.

[0073] S108, Construct a wear mapping library: Construct a wear mapping library containing wear states and characteristic quantities based on historical data, including decoupled characteristic benchmarks for normal, light, and heavy wear;

[0074] S109, Wear State Identification: The wear features of a single tool are compared with the mapping library to identify the wear state. The wear state of each tool is initially determined by cosine similarity calculation, thereby generating the preliminary wear state of each tool as an anchoring reference to lock the approximate range to which the wear features of the tool belong.

[0075] S110, Dynamically adjust the classification threshold: The statistical distribution of feature values ​​is calculated using a sliding time window, and the classification threshold is dynamically adjusted based on the standard deviation of the distribution and the interaction compensation matrix.

[0076] S111, whether the feature value continues to exceed the classification threshold: to correct the judgment criteria, once the feature value of the target tool continues to exceed the classification threshold, S112 is executed;

[0077] S112, marked as abnormal wear;

[0078] S113, triggers the tool change mechanism;

[0079] S114, Generate Tool Changing Sequence: While generating the tool changing sequence, output the tool barrel load balance parameter adjustment command to reduce the load in the area where the tool has not been changed, avoid the decrease in construction efficiency and safety risks caused by inaccurate positioning or delayed decision-making, and suppress the wear chain reaction.

[0080] In this embodiment, in the actual operation scenario of the tunnel boring machine cutter barrel, there are various types of cutters with significant functional differences. The cutter barrel is equipped not only with roller cutters for breaking hard rock and bearing high impact loads, but also with scrapers for scraping soft soil and bearing continuous friction. The interaction mechanism, stress characteristics, and wear patterns of the two types of cutters with the rock and soil are fundamentally different. Relying on the independent laser vibration measurement device behind each cutter, the vibration signal of a single cutter can be directly acquired without signal separation processing. Then, the high-frequency impact of the roller cutter and the low-frequency vibration characteristics of the scraper are captured by multi-scale time-frequency transformation. Combined with infrared thermal imaging, the temperature change patterns of the two types of cutters are distinguished. The roller cutter is prone to local high temperature points due to impact and friction, while the scraper shows an overall temperature rise. The load transfer across different types of cutters is quantified by an interactive compensation matrix, ultimately achieving accurate wear monitoring under complex cutter combinations and solving the problem of signal analysis and interactive effect processing caused by type differences.

[0081] Specifically, the multi-scale feature extraction process is as follows: Figure 2 As shown, the specific steps include S201-221:

[0082] S201, Obtain tool physical parameters: Based on the tool head design drawings, obtain the physical parameters of all tools, including the theoretical coordinates, tool type label and unique number of each tool;

[0083] S202, Construct a tool type library: Identify each tool using key-value pairs to build the tool type library. Tool types include, but are not limited to, center hobs, transition hobs, edge hobs, front scrapers, and edge scrapers. Numbers are used to distinguish each tool, such as... Indicates the first in the cutter head The first of the categories of cutting tools One knife, , This refers to the number of different types of cutting tools in the cutter head. , For the first The number of tools in each category of cutting tools;

[0084] S203, Classification of Wear Risk: Based on historical wear data, the average wear rate of various types of tools is extracted, and the wear risk of tool types is classified into Level 1 risk, Level 2 risk, and Level 3 risk.

[0085] Regarding the above S203: the tools that come into contact with the edge of the working face are marked as Level 1 risk, such as edge roller cutters and edge scrapers. Because they come into contact with the tunnel wall and complex strata, their wear rate is 2-3 times that of the central tools. Tools in the transition area are marked as Level 2 risk, such as transition roller cutters and front scrapers. Tools in the central area are marked as Level 3 risk, such as the central roller cutter.

[0086] S204, Generate Scan List: Set the corresponding scan priority and scan interval, generate the scan list, and associate it with the tool number in real time;

[0087] The scanning priority order corresponding to S204 above is level one, level two, and level three risk, with corresponding scanning intervals of... , , And satisfy as ;

[0088] S205, Equipment Time Calibration: Based on the independent laser vibration measuring device behind each tool, a laser beam is emitted to scan the target tool, and time calibration is performed at the same time to ensure that the timestamp of the vibration measuring device is completely synchronized with the timestamp of the infrared thermal imaging device.

[0089] S206, Generate Scan Command: Generates a single-blade scan command, including scan priority, sampling frequency, and duration, to ensure the spatiotemporal matching of vibration signals and temperature data;

[0090] S207, Collect mixed signals: Collect the reflected signal of the target tool through the laser receiver, which includes the vibration information of the target tool, local dust scattering noise and the inherent electromagnetic interference of the equipment;

[0091] S208, Noise Filtering: Combined with a narrowband bandpass filter, non-target frequency noise is filtered to obtain a preliminary filtered signal;

[0092] S209, generating vibration signal: An adaptive noise canceller is constructed through the least mean square algorithm to iteratively suppress dust and electromagnetic interference, and finally generate a target tool vibration signal with high signal-to-noise ratio;

[0093] Regarding the above 209: Since the vibration signal of a single blade is a continuous time series, even if the wear state is stable, there are still differences in dynamic characteristics. In order to avoid information loss caused by processing the whole segment, the vibration signal is processed in segments.

[0094] S210, Calculate target variance: Calculate the target variance of the vibration signal, and based on the stage parameter library and combined with the tool type, call the wear threshold of the target tool;

[0095] S211, Obtain the wear threshold;

[0096] S212, Is the target variance less than the wear threshold? If the target variance is less than the wear threshold, it is determined to be a stable stage. If the target variance exceeds the wear threshold, it is determined to be a rapid stage. Thus, a wear stage identifier for the vibration signal is generated, including stable and rapid stages.

[0097] The rate of change of signal characteristics varies at different wear stages, requiring matching of segment lengths and noise reduction intensities to ensure feature integrity.

[0098] S213, Call segment duration: Based on the stage parameter library, call the segment duration of the corresponding stage to extract the corresponding segment signal;

[0099] S214, generating vibration segmented signals: combining wavelet thresholding for noise reduction, a general threshold is used in the stable stage to preserve the trend, and an attenuation coefficient is introduced in the abrupt stage, combined with the general threshold to preserve the abrupt changes, finally generating the vibration segmented signals of the target tool.

[0100] S215, Generate low-frequency matrix: For the vibration segmented signal, call the STFT parameter library to extract low-frequency features, thereby generating a low-frequency matrix;

[0101] S216, Generate high-frequency matrix: Call the wavelet parameter library, configure the corresponding decomposition parameters, extract high-frequency band features, and thus generate a high-frequency matrix;

[0102] S217, Calculate the global difference: Based on the calculation of standard deviation, combined with the low-frequency matrix and high-frequency matrix, calculate the low-frequency difference and high-frequency difference to reflect the characteristic fluctuations, and obtain the global difference by the sum of the low-frequency difference and high-frequency difference;

[0103] S218, Generate low-frequency contribution and high-frequency contribution: Calculate the low-frequency contribution based on the ratio of low-frequency difference to global difference, and similarly calculate the high-frequency contribution. At this time, the sum of low-frequency and high-frequency contributions is 1.

[0104] S219, Extract correction coefficients: Call the weight correction library and extract the corresponding correction coefficients based on the wear stage identifier of the target tool;

[0105] S220, Calculate low-frequency weight and high-frequency weight: For the stable stage, the low-frequency weight is calculated by the correction coefficient and the low-frequency contribution. Since the sum of the low-frequency weight and the high-frequency weight is 1, the high-frequency weight is also calculated. For the rapid stage, the high-frequency weight and the corresponding low-frequency weight are calculated by the correction coefficient and the high-frequency contribution.

[0106] S221, Generate time-frequency energy spectrum: Based on the corresponding weights, combine the low-frequency matrix and the high-frequency matrix, perform weighted calculations, and fuse them to generate a time-frequency energy spectrum.

[0107] Specifically, the process for identifying localized overheating areas is as follows: Figure 3As shown, the specific steps include S301-313:

[0108] S301, Acquire temperature field image: Since the local overheating of the tool is closely related to the abnormal vibration, when the wear intensifies, the vibration intensity of the tool and the frictional heat generation will rise simultaneously. In order to achieve the spatiotemporal matching of the two types of data, an infrared thermal imager is deployed in the non-rotating area of ​​the tool barrel to collect the temperature field image of the entire tool barrel in real time. The timestamps of the thermal imager and the laser vibration meter are synchronized to ensure that each frame of temperature field image corresponds to the vibration signal at the same moment.

[0109] The acquired raw temperature field image is easily affected by dust scattering and water mist interference, resulting in noise bright spots or blurred areas. At this time, the temperature field image is preprocessed. Based on smoothing the image with Gaussian filtering, high-frequency noise such as bright spots caused by dust scattering is removed. Adaptive histogram equalization is used to enhance the temperature contrast between the cutting edge and the background, making the temperature difference between the cutting edge and the tool body and the background more obvious.

[0110] S302, Calculate pixel gradient: Tool wear mainly occurs at the cutting edge where it is in direct contact with the soil and rock. Its temperature change can most directly reflect the wear state. However, the temperature of other areas of the tool body is greatly affected by heat dissipation and has low reference value. Edge detection is performed on the preprocessed temperature field image, and the pixel gradient is calculated using the Sobel operator to obtain the gradient magnitude matrix and gradient direction matrix. Local maxima along the gradient direction are retained by non-maximum suppression. Each pixel in the image is traversed to determine whether the pixel is the maximum value in its neighborhood. If it is the maximum value, the current pixel is retained. Otherwise, the gradient magnitude of the current pixel is modified to 0 to suppress non-edge pixels.

[0111] S303, set dual thresholds: refine the edge to a single pixel width, highlight the continuous contour of the blade, and set dual thresholds, including a first threshold and a second threshold. Pixels with gradient magnitude greater than the first threshold are defined as strong edges, pixels with gradient magnitude less than the second threshold are defined as non-edges, and pixels with gradient magnitude between the second threshold and the first threshold are defined as weak edges. Traverse the strong edge pixels and mark their positions. When the weak edge pixels and the strong edge pixels are within a preset neighborhood, they are determined to be real edges and their positions are marked. Otherwise, they are noise and are suppressed, thereby generating an edge image that only contains the edge contour.

[0112] S304, Generate edge image: The neighborhood of a pixel is the left and right adjacent pixels along the gradient direction of the pixel, and the adjacent pixel is the pixel closest to the target pixel;

[0113] S305, Hough Transform: Performs Hough transform on the edge image to extract the pixel coordinates of all continuous edges;

[0114] S306, Generate a pixel set for the cutting edge region: Combine the actual coordinates in the tool type library, filter out edge segments within the preset range of the tool's theoretical coordinates, retain continuous edges whose length conforms to the characteristics of the tool's cutting edge, thereby generating a pixel set for the cutting edge region to accurately segment the cutting edge region;

[0115] S307, Calculate pixel temperature gradient: The temperature gradient of the cutting edge reflects the degree of local wear better than the overall temperature. During normal wear, the heat distribution is relatively uniform and the gradient is small.

[0116] S308, Obtain the local temperature gradient of the cutting edge: When abnormal wear occurs on the cutting edge, local friction intensifies, heat concentrates, and the gradient will increase significantly. Based on the pixel coordinates of the cutting edge area, the temperature value of each pixel is extracted from the temperature field image. Adjacent pixels are selected along the direction perpendicular to the cutting edge, and the ratio of the temperature difference between the two points to the actual physical distance is calculated to obtain the temperature gradient of the corresponding pixel coordinates.

[0117] S309, Get the upper limit of gradient: Take the maximum temperature gradient value in the cutting edge area as the local temperature gradient of the cutting edge of the target tool;

[0118] S310, Set dynamic coefficient: The normal temperature gradient of different tools varies due to working conditions, and normal vibration and shock can also cause a short-term temperature rise. If a fixed threshold is used, misjudgment is likely to occur. Call the normal temperature gradient range of similar tools in the historical database, including the lower limit and upper limit of the gradient, and obtain the vibration amplitude of the target tool, and set the dynamic coefficient.

[0119] S311, Calculate the over-temperature threshold: The over-temperature threshold is obtained by multiplying the vibration amplitude and the dynamic coefficient, and adding the upper limit of the gradient. The threshold is then corrected. The larger the vibration amplitude, the higher the allowable normal gradient threshold.

[0120] S312, Filtering local overheated areas: When the local temperature gradient of the cutting edge is greater than the overheating threshold and the duration exceeds the preset time interval, it is determined to be a local overheated area;

[0121] S313, Construct an over-temperature list: Convert the pixel coordinates of the cutting edge in the local over-temperature area into actual spatial coordinates through perspective transformation, associate them with the tool number, and generate an over-temperature list that includes tool changes, actual cutting edge position, temperature gradient value, and over-temperature duration.

[0122] Specifically, the wear and tear interaction resolution process is as follows: Figure 4 As shown, the specific steps include S401-421:

[0123] S401, Construct a multi-dimensional feature set: Integrate the time-frequency energy spectrum of the target tool, the local temperature gradient of the cutting edge, and the over-temperature marker to construct a multi-dimensional feature set of the target tool; among which, the over-temperature marker includes over-temperature and normal.

[0124] S402, Generate target neighborhood set: Calculate the Euclidean distance between the target tool and other tools in the tool holder, set a neighborhood threshold, and define tools whose Euclidean distance is less than the neighborhood threshold as neighboring tools of the target tool, thereby generating the target neighborhood set. , For the target tool Take the neighbor's knife, , Number of adjacent blades;

[0125] S403, Constructing Adjacent Tool Interaction Features: The influence of adjacent tools on the target tool is not linearly additive. For example, when two adjacent tools wear simultaneously, the load on the target tool increases non-linearly, making it difficult for linear models to depict the real interaction. For both the target tool and adjacent tools, based on the adjacent tools... The high-frequency matrix is ​​inversely proportional to the square of the distance, the product of the temperature gradient and the thermal conductivity coefficient of the tool barrel, and the multidimensional features corresponding to neighboring tools are used to construct the neighboring tool interaction features. This reflects the combined effects of vibration, proximity, and heat transfer. The number of features;

[0126] S404, Constructing a correlation model: Using the actual wear of the target tool as the dependent variable, the multidimensional feature set of the target tool and the interaction features of adjacent tools as independent variables, a multiple linear regression with interaction terms is adopted, and historical data is used for training to generate a correlation model;

[0127] S405, Generate Predicted Wear: The correlation model outputs the predicted wear of the target tool;

[0128] S406, set the secondary wear threshold;

[0129] S407, Classify wear condition: Classify the wear condition of the target tool, including normal, light, and heavy wear, and mark the wear condition.

[0130] S408, Obtain the interaction coefficient: based on any adjacent tool of the target tool. Obtain the neighboring knife in the association model Interaction coefficient ;

[0131] S409, Calculate the weight of each feature: Normalize the interaction coefficients of neighboring tools to obtain the weight of each feature in the neighboring tool interaction features;

[0132] S410, Calculate the compensation factor for adjacent tools: distinguish between adjacent tools. The influence intensity of different features on the target tool is determined, and a weighted calculation is performed on each feature to obtain the adjacent tool. Compensation factor for the target tool;

[0133] S411, Generate Interaction Compensation Matrix: Integrate the compensation factors of each neighboring tool in the target neighborhood set to generate the interaction compensation matrix of the target tool; For S411, the value is 0 for neighboring tools that are not the target tool or for the corresponding position of the target tool.

[0134] S412, Construct an interactive topology network: Take all the tools on the tool barrel as network nodes, each node corresponds to a unique tool number, assign node attributes to each node, including wear intensity and node degree. Wear intensity is the output of the association model about the target tool, and node degree is the number of neighboring tools of the target tool. The neighboring tool pairs formed by the target tool and the neighboring tools are used as edges. Tools that are not neighboring tools or have no influence are not connected to each other to reduce network redundancy. The corresponding compensation factor is the weight of the edge. Directed arrows are used to mark the direction of influence from the neighboring tools to the target tool to construct the interactive topology network of the target tool.

[0135] S413, Calculate node influence: Combine the compensation factor and use the PageRank algorithm to calculate node influence;

[0136] S414, Filtering key interference sources: Select nodes with influence greater than the influence threshold as key interference sources. For the target tool, start from the key interference source, trace the directed edges with weight greater than the interference threshold to form key interference paths, and mark them.

[0137] S415, Calculate the interference components of adjacent tools: For the target tool, only retain the adjacent tools on the critical interference path and ignore the non-critical adjacent tools. If the weight is not greater than the interference threshold or is not on the critical path, reduce invalid corrections. The vibration of the adjacent tools will be transmitted to the target tool through the tool barrel structure, thus affecting the target tool. Based on the dynamic model of the tool barrel structure, calculate the attenuation coefficient. At the same time, combine the time-frequency energy spectrum of the adjacent tools and the compensation factor to perform a product operation to obtain the interference components of a single adjacent tool.

[0138] S416, obtain the quantitative interference amount: accumulate the interference components of all adjacent tools of the target tool to obtain the quantitative interference amount of the target tool, which reflects the actual interference intensity transmitted by the structure;

[0139] S417, Generate actual time-frequency characteristics: Remove the directional interference from the time-frequency energy map of the target tool to obtain the actual time-frequency characteristics of the target tool. If the actual time-frequency characteristics are negative, it means that the interference has completely covered the original signal, and the actual time-frequency characteristics are changed to 0.

[0140] S418, Calculate feature fidelity: Calculate feature fidelity based on the degree of difference in low-frequency trend features before and after decoupling;

[0141] S419, Is the feature fidelity less than the decoupling threshold? If the feature fidelity is less than the decoupling threshold, it is determined that the decoupling has not excessively damaged its own features; otherwise, the weight of the corresponding path is reduced and recalculated.

[0142] S420, Feature Extraction: Extract core features from the decoupled spectrum, such as low-frequency trend slope, high-frequency energy peak, and characteristic frequency bandwidth.

[0143] S421, Construct Wear Features: Form the wear features of the target tool, associate the tool number with the timestamp, and generate a lookup table containing decoupling features and wear status.

[0144] Specifically, the steps for wear condition identification include:

[0145] To avoid identification errors caused by differences in features among different types of cutting tools, the wear features of all cutting tools after decoupling are extracted from the historical database, grouped by cutting tool type, and the K-means clustering algorithm is used to perform cluster analysis on wear features of the same type and state. The center vector of each feature is taken as the reference feature of this type-state, and the normal fluctuation range of each feature is recorded. The reference vector and the corresponding feature fluctuation range are stored as key-value pairs of "type-state-reference feature", forming a type-based wear mapping library.

[0146] Since the absolute values ​​of the characteristics of different individual cutting tools vary, and the relative trends of change between characteristics are a better indicator of wear status, based on the wear characteristics of the target cutting tool and determining its type, three baseline features corresponding to the type are retrieved from the wear mapping library. , , Calculate the cosine similarity between the wear characteristics of the target tool and the three reference characteristics. , , ,in, , , These are the baseline characteristics corresponding to normal, light, and heavy wear conditions, respectively. , , The target tool and , , The cosine similarity is used to determine the matching degree by measuring the consistency of the direction of the feature vectors. The closer the value is to 1, the more consistent the trend of the current feature with the baseline feature is, and the higher the matching degree.

[0147] The similarity scores between the target tool and three benchmark features are ranked, and the maximum value is selected. The corresponding wear state is used as the anchoring state of the target tool. This generates preliminary judgment results that include anchoring status and similarity distribution;

[0148] Since the wear characteristics of the tool are affected by the real-time working conditions, such as the normal vibration characteristics in a hard rock environment may be similar to the slight wear characteristics in a soft soil environment, real-time working condition parameters are obtained, and the working condition correction factor is calculated based on the difference between the current working condition and the standard working condition. The benchmark characteristics of the anchoring state are then corrected to form a working condition adaptation benchmark.

[0149] Calculate the similarity between the target tool wear characteristics and the working condition adaptation benchmark. If the similarity increases and the working condition adaptation benchmark remains unchanged, the similarity is considered to be positive. If the wear condition is within the normal fluctuation range, the original anchoring state is maintained; otherwise, it indicates that the original anchoring state does not match the current working condition. In this case, the benchmark feature matching and screening are performed again to finally generate the wear state after working condition correction, which serves as the basis for tool replacement decision.

[0150] Specifically, the steps for generating the tool change sequence include:

[0151] Tools with light or heavy wear are identified and defined as worn tools. Wear features of worn tools are extracted. To avoid interference from neighboring tools, the classification threshold corresponding to the worn tool is called. At the same time, based on the interaction compensation matrix, the compensation factors of all neighboring tools of the worn tool are extracted, summed, and the neighboring tool interference weight is obtained. The classification threshold is then corrected to obtain the anomaly judgment threshold.

[0152] Based on the wear characteristics of the worn tool, the latest real-time feature value is called to determine whether it exceeds the anomaly judgment threshold; if the real-time feature value is less than the anomaly judgment threshold, it is judged as a mild anomaly; otherwise, it is judged as a severe anomaly.

[0153] Perform duration verification; for minor anomalies, if continuous... If the real-time feature value within a window exceeds the anomaly detection threshold, it is judged as primary abnormal wear. For severe anomalies, if the value is continuously... If the real-time feature value within a window exceeds the anomaly judgment threshold, it is judged as an emergency abnormal wear. At the same time, the worn tool with a severe wear status is marked as a near-term abnormal wear. Finally, an abnormal wear tool list is generated, including tool number, anomaly type, and first time exceeding the threshold. The anomaly type includes near-term, primary, and emergency.

[0154] Tool replacement must take into account both priority and load balance. It cannot be simply sorted by wear level. Priority should be assigned to abnormally worn tools, with emergency abnormal wear having the highest priority and needing to be replaced in a short time. Primary abnormal wear is the next highest priority, and near-expiration abnormal wear has the lowest priority.

[0155] Based on the spatial coordinates of the tool and the interaction compensation matrix, the positional relationship of the candidate tool changers is verified. If the distance between two tools is less than the load distance threshold, they cannot be included in the same batch of tool changers.

[0156] A greedy algorithm incorporating real-time load feedback is used to generate a tool change sequence. First, the highest priority tool is selected. Then, based on the real-time load characteristics measured directly from vibration, high-priority tools whose vibration amplitude exceeds the safety threshold among their close neighbors are eliminated. Next, the second highest priority tool is selected from the remaining tools, and so on. At the same time, it is ensured that the number of tools changed in each batch does not exceed the capacity limit of the tool changer. Finally, a batch-based tool change sequence is formed, specifying the tool ID and the latest replacement time for each batch.

[0157] During tool changing, the unreplaced tool bears an additional load. If the load is too high, it will accelerate its wear. It is necessary to reduce the load on the unreplaced tool area by adjusting the parameters to suppress the chain reaction. Calculate the original load ratio of the tool changing area and combine it with the reserved safety redundancy load to calculate the load redundancy of the unreplaced tool area, which is the upper limit of the additional load that the unreplaced area can withstand. Generate regional adjustment instructions, such as reducing the feed speed or tool barrel pressure of the unreplaced tool area, to ensure that its load does not exceed the safe range. At the same time, the adjustment instructions must be associated with the time node of the tool changing sequence. They take effect when the tool changing begins and return to normal parameters after the change is completed, so as to reduce the impact on construction efficiency while ensuring safety.

[0158] Example 2

[0159] Another embodiment of the present invention provides: a wear monitoring system for tunnel boring machine cutters, comprising: a separation module, a monitoring module, a wear analysis module, and a cutter replacement identification module;

[0160] The separation module is used to scan the target tool by setting an independent laser vibration measurement device behind each tool inside the tool barrel. Combined with dynamic coordinate correction and orthogonal verification to filter non-target signals, it achieves the initial separation of vibration signals of multiple tools. At the same time, it superimposes narrowband optical filtering and lock-in amplification technology to suppress noise, thereby avoiding signal coupling and aliasing of multiple tools working together from the root and ensuring the independent extraction of vibration signals of a single tool.

[0161] The monitoring module is used to extract multi-scale features based on the vibration signal acquired by the laser vibrometer. It extracts low-frequency trend features through STFT and high-frequency impact features through wavelet packet decomposition, and generates a time-frequency energy spectrum to capture multi-scale wear features. Simultaneously, it combines infrared thermal imaging to acquire the temperature field of the tool barrel, and uses edge detection and Hough transform to identify the tool cutting edge area, extract local temperature gradients, identify local overheating areas, and associate them with the corresponding tools. This achieves dual-dimensional monitoring including vibration and temperature, comprehensively capturing the dynamic features and thermal response of tool wear, and avoiding the omission of complex wear modes by a single feature.

[0162] The wear analysis module is used to establish a mapping relationship based on time-frequency energy spectrum and temperature gradient data, combined with historical wear records. It calculates the inter-tool compensation factor and generates an interactive compensation matrix through multivariate regression analysis to quantify the nonlinear influence of adjacent tool wear. At the same time, it constructs a topology network with tools as nodes, corrects the time-frequency energy spectrum according to the compensation factor, and removes interference from adjacent tools to restore the true wear characteristics of a single tool.

[0163] The tool change identification module utilizes a categorized wear mapping library to accurately identify normal, light, and heavy tool wear states through cosine similarity calculation and adaptive correction based on working conditions. This addresses the issue of individual tool wear characteristics being masked due to signal distortion. Combined with an interactive compensation matrix, the module dynamically adjusts the classification threshold, marking tools that consistently exceed the threshold as having abnormal wear. Tool change priorities are assigned based on the abnormality type, and a batch tool change sequence is generated by combining tool spatial coordinates and load balance constraints. Simultaneously, it outputs instructions to adjust the tool barrel load balance parameters to reduce the load on areas where tools have not been changed, avoiding construction risks caused by inaccurate positioning or delayed decision-making. This effectively suppresses wear chain reactions and ensures the accuracy of tool change decisions and construction safety.

[0164] Working principle and effects:

[0165] The target tool is scanned by an independent laser vibration measurement device, and dynamic coordinate correction is used to ensure the accuracy of laser scanning. The coupling signals of multiple tools are separated from the source, and the accuracy of single tool feature separation is significantly improved. Vibration features are extracted by multi-scale time-frequency transformation, and infrared thermal imaging and cutting edge temperature gradient analysis are combined to construct a two-dimensional feature set. For the interaction effect of multiple tools, the nonlinear influence of load transmission between adjacent tools is quantified by the correlation model. The key interference path is identified by combining topological network, and the interference of adjacent tools is selectively eliminated to restore the true characteristics of single tools. This solves the problem of contamination of healthy tool signals and masking of worn tool signals, and the wear positioning accuracy is greatly improved. After the state recognition is realized based on the wear classification mapping library and working condition adaptive correction, the tool replacement sequence is generated according to priority and load balance. The load of the non-replaced tool area is adjusted synchronously, which reduces the misjudgment rate and the risk of chain wear. This ensures construction safety and improves the efficiency of tool replacement decision-making.

[0166] The above description is merely a preferred embodiment of the present invention. The scope of protection of the present invention is not limited to the above embodiments. All technical solutions falling within the scope of the present invention's concept are within the scope of protection of the present invention. It should be noted that for those skilled in the art, any improvements and modifications made without departing from the principles of the present invention should also be considered within the scope of protection of the present invention.

Claims

1. A method for monitoring the wear of tunnel boring machine cutters, characterized in that, include: A laser vibration measuring device is installed behind the tool inside the tool barrel to scan the target tool, obtain the vibration signal of the target tool, and perform multi-scale feature extraction to generate a time-frequency energy spectrum. The temperature field distribution of the tool barrel is obtained, local overheating areas are identified, and associated with the corresponding tool coordinates. Based on the time-frequency energy spectrum and temperature gradient data, combined with historical wear records, a mapping relationship including wear, temperature, and vibration is established. Wear interaction analysis is performed, and the interaction compensation matrix is ​​calculated. At the same time, a tool topology graph is constructed with the tool as a node. The interaction compensation matrix is ​​used as the weight of the edge to perform weighted correction on the time-frequency energy spectrum, eliminating the interference of adjacent tool wear on the current tool vibration signal, thereby generating the wear characteristics of the target tool. Based on historical data, a wear mapping library containing wear states and feature quantities is constructed. The wear features of individual tools are compared with the mapping library, and the wear state of each tool is initially judged by cosine similarity calculation to identify the wear state. This generates the preliminary wear state of each tool, and the classification threshold is dynamically adjusted. Once the feature value of the target tool continuously exceeds the classification threshold, it is marked as abnormal wear, triggering the tool change mechanism and generating a tool change sequence.

2. The method for monitoring wear of tunnel boring machine cutters according to claim 1, characterized in that, The steps of multi-scale feature extraction include: Obtain the physical parameters of each tool in the tool barrel and generate a tool type library; Based on historical wear data, wear risk is classified according to tool type, and corresponding scanning priority and scanning interval are set to generate a scanning list; The laser vibration measurement device is aligned with the target tool, and the time of each device is calibrated to generate a single-tool scanning command. The reflection signal of the target tool is collected, noise is filtered, and a preliminary filtered signal is obtained. A noise canceller is constructed using the least mean square algorithm to obtain the vibration signal of the target tool.

3. The method for monitoring wear of tunnel boring machine cutters according to claim 2, characterized in that, The steps of multi-scale feature extraction also include: Calculate the target variance of the vibration signal, obtain the wear threshold, if the target variance is less than the wear threshold, it is a stable stage; otherwise, it is a rapid stage, and generate a wear stage identifier. Based on the stage parameter library, the segment duration of the corresponding stage is called, the segment signal is extracted, and wavelet threshold denoising is combined to generate the vibration segment signal of the target tool. The STFT parameter library is called to generate a low-frequency matrix, and the wavelet parameter library is called to configure the corresponding decomposition parameters, thereby generating a high-frequency matrix. Calculate the low-frequency difference and high-frequency difference to obtain the global difference, thereby generating the low-frequency contribution and high-frequency contribution. The weight correction library is called to extract the correction coefficients, calculate the low-frequency weights and high-frequency weights, and perform weighted calculations to generate a time-frequency energy spectrum.

4. The method for monitoring wear of tunnel boring machine cutters according to claim 3, characterized in that, The steps to identify localized overheating areas include: The temperature field image of the cutter barrel is acquired in real time and preprocessed. The Sobel operator is used to calculate the pixel gradient. Local maxima along the gradient direction are preserved by non-maximum suppression. Double thresholds are set to distinguish strong edges, weak edges and non-edges, and an edge image is generated. Perform a Hough transform on the edge image to generate a set of pixels for the blade edge region; Calculate the temperature gradient of each pixel, and take the maximum temperature gradient value in the cutting edge area as the local temperature gradient of the cutting edge of the target tool. Obtain the upper limit of the normal temperature gradient, set the dynamic coefficient, and calculate the over-temperature threshold by combining it with the vibration amplitude of the target tool; If the local temperature gradient of the cutting edge is greater than the over-temperature threshold and the duration exceeds a preset time interval, it is determined to be a local over-temperature region, and actual spatial coordinates are generated to construct an over-temperature list.

5. The method for monitoring wear of tunnel boring machine cutters according to claim 4, characterized in that, The steps involved in wear and tear interaction parsing include: Construct a multidimensional feature set of the target tool, calculate the Euclidean distance between the target tool and other tools in the tool barrel, combine the neighborhood threshold to generate a target neighborhood set, and construct neighbor tool interaction features; Using the actual wear amount of the target tool as the dependent variable and the multidimensional feature set and the neighbor tool interaction features as independent variables, an association model is constructed to generate the predicted wear amount of the target tool. Set a two-level wear threshold to classify the wear state of the target tool, including normal, light, and heavy. Based on any neighboring tool of the target tool, obtain the interaction coefficient with the neighboring tool in the association model, and perform normalization processing to obtain the weight of each feature in the neighboring tool interaction features; Each feature in the neighboring tool interaction features is weighted to obtain the compensation factor of the neighboring tool, and combined with the target neighborhood set to generate the interaction compensation matrix of the target tool.

6. The method for monitoring wear of tunnel boring machine cutters according to claim 5, characterized in that, The steps of wear and tear interaction parsing also include: Using the cutting tools on the tool barrel as network nodes, each node is assigned node attributes, including wear intensity and node degree. Adjacent tool pairs are used as edges, and the corresponding compensation factors are used as edge weights to construct the interactive topology network of the target tool. By combining compensation factors, the PageRank algorithm is used to calculate node influence, and key interference sources are screened out by combining influence thresholds, and key interference paths are marked. For the target tool, only the adjacent tools on the critical interference path are retained, the interference component of a single adjacent tool is calculated, and the components are accumulated based on the adjacent tools to obtain the quantitative interference amount of the target tool. Remove the directional interference from the time-frequency energy map of the target tool to obtain the actual time-frequency characteristics, and calculate the feature fidelity. Once the feature fidelity is not less than the decoupling threshold, recalculate the path weight. For the decoupled image, features are extracted to construct the wear characteristics of the target tool.

7. The method for monitoring wear of tunnel boring machine cutters according to claim 6, characterized in that, The steps for wear condition identification include: The historical wear characteristics of each tool in the tool barrel are obtained, and cluster analysis is performed on the wear characteristics of the same type and state to obtain the baseline characteristics, thereby generating a wear mapping library; Based on the wear characteristics of the target tool, reference features are invoked, and the similarity between the wear characteristics and the reference features is calculated. The wear state corresponding to the maximum similarity is selected as the anchoring state. Real-time operating parameters are acquired, operating condition correction factors are calculated, and the benchmark features of the anchoring state are corrected to form an operating condition adaptation benchmark. The similarity between the damage features and the operating condition adaptation benchmark is calculated. If the similarity is improved and the working condition adaptation benchmark is still within the normal fluctuation range of the anchoring state, the original anchoring state is maintained; otherwise, the benchmark feature matching and screening are performed again.

8. The method for monitoring wear of tunnel boring machine cutters according to claim 7, characterized in that, The steps for generating the tool change sequence include: The worn tool and its corresponding classification threshold are obtained. Based on the compensation factor of the worn tool, the interference weight of the adjacent tool is calculated, and the classification threshold is corrected to obtain the anomaly judgment threshold. The real-time characteristic value of the worn tool is obtained. If the real-time characteristic value is less than the anomaly judgment threshold, it is judged as a slight anomaly; otherwise, it is judged as a severe anomaly. For mild abnormalities, if continuous If the real-time feature value within a window exceeds the anomaly detection threshold, it is judged as primary abnormal wear. For severe anomalies, if the value is continuously... If the real-time feature value in a window exceeds the anomaly judgment threshold, it is judged as an emergency abnormal wear. At the same time, the worn tool with a severe wear condition is marked as an imminent abnormal wear, and finally a list of abnormal wear tools is generated. in, The number of decision windows required to determine primary abnormal wear. To determine the number of decision windows required for emergency abnormal wear, , All settings are based on statistical verification of historical data, and ; Prioritize abnormally worn tools, verify the position of candidate tool changers, and generate tool change sequences using a greedy algorithm; During tool change, the original load percentage of the tool change area is calculated, and combined with the reserved safety redundancy load, the load redundancy of the non-tool change area is calculated to generate regional adjustment commands.

9. A wear monitoring system for tunnel boring machine (TBM) cutters, used to implement the wear monitoring method for TBM cutters as described in any one of claims 1-8, characterized in that, include: Separation module, monitoring module, wear analysis module, and tool change identification module; The separation module is used to scan the target tool and obtain vibration signals by setting a laser vibration measuring device behind the tool inside the tool barrel; The monitoring module is used to construct a time-frequency energy spectrum and collect the temperature field of the cutter barrel in real time to identify local overheating areas; The wear analysis module is used to construct wear mapping relationships, generate interactive compensation matrices, construct a topology network with the tool as nodes, and correct the time-frequency energy spectrum. The tool change identification module is used to initially determine the wear status, dynamically adjust the classification threshold in combination with the interactive compensation matrix, identify abnormally worn tools, and generate a batch tool change sequence.

Images