Tunnel sprayed concrete quality intelligent detection method based on multi-sensor fusion

Abstract

The invention discloses a tunnel shotcrete quality intelligent detection method based on multi-sensor fusion, and relates to the technical field of tunnel engineering, and the method comprises the following steps: S1, constructing a spatial modeling process of a thermal radiation area of shotcrete equipment, and according to heat conduction paths and thermal radiation duration time parameters of the shotcrete equipment in different operation states, calculating the thermal radiation duration time parameters of the shotcrete equipment; and establishing a thermal interference influence area prediction model updated along with time, and generating a thermal interference area prediction map. According to the method, misjudgment caused by thermal interference of equipment is effectively eliminated through thermal radiation area modeling, image registration and dynamic mask shielding technologies; according to the method, hot spot features and time sequence features are combined, a confidence label is generated by using a discriminant learning model, accurate optimization and feedback control of spraying parameters are realized, an infrared detection and equipment regulation and control closed-loop system is constructed, intelligence, precision and stability of sprayed concrete quality control are remarkably improved, and material waste and repair cost are reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel engineering, and particularly relates to a tunnel shotcrete quality intelligent detection method based on multi-sensor fusion. BACKGROUND

[0002] The tunnel shotcrete quality intelligent detection based on multi-sensor fusion refers to, in the tunnel construction process, comprehensively using multiple types of sensors (such as high-definition cameras, laser scanners, ultrasonic sensors, infrared thermal imagers and temperature and humidity sensors) to synchronously and multi-angulately collect data on the surface state, internal structure and construction environment of the shotcrete, and through a unified interface, accessing a data processing platform, using a machine learning algorithm to fuse, analyze and abnormally identify multi-source heterogeneous data, realizing real-time detection and accurate evaluation of concrete defects (such as cracks, hollowing, honeycombing, temperature control abnormalities and the like). The method not only overcomes the limitations of poor adaptability to complex environments and low detection precision of traditional single detection means, but also can upload analysis results to the cloud through a wireless communication module and automatically feed back to the shotcrete equipment for parameter adjustment, thereby building an intelligent detection system with a “perception-analysis-feedback” closed loop capability, significantly improving the intelligentization and refinement level of tunnel shotcrete quality control.

[0003] The prior art has the following disadvantages: in the process of shotcrete quality detection based on infrared sensors, the prior art usually relies on infrared thermal imaging equipment to monitor the temperature distribution on the surface of the concrete in real time, so as to identify areas of insufficient hydration or internal structural defects such as hollowing and cracking. However, in the process of shotcrete construction, if the shotcrete equipment is in an intermittent operation state, the heat existing on the surface of the equipment will continue to radiate to the surrounding concrete area, easily forming a local non-uniform heat diffusion zone near the equipment. Since such heat anomalies are not caused by structural problems of the concrete itself, but by equipment heat radiation, and their thermal imaging features are highly similar to the heat spot patterns of real hollowing or unhydrated areas in the image, they are easily misidentified as quality defects by the detection system. If the system automatically triggers feedback control instructions such as “re-shotcrete” or “adjust the shotcrete angle” based on such misjudgment results, it may cause the qualified areas to be repeatedly worked on, thereby causing problems such as local over-shooting of the shotcrete, material accumulation or surface damage, affecting the forming quality and mechanical properties of the shotcrete layer, increasing material consumption and subsequent repair costs, and even seriously interfering with the stability of the overall supporting structure.

[0004] The above information disclosed in the background section is only used to enhance the understanding of the background of the present disclosure, and therefore it can include information that does not constitute prior art known to those of ordinary skill in the art. SUMMARY

[0005] The purpose of the present application is to provide a tunnel shotcrete quality intelligent detection method based on multi-sensor fusion, which effectively eliminates misjudgment caused by equipment thermal interference through thermal radiation area modeling, image registration and dynamic mask shielding technology; combines thermal spot features and time sequence features, generates confidence labels using a discriminative learning model to realize precise optimization and feedback control of the spraying parameters, constructs an infrared detection and equipment control closed-loop system, significantly improves the intelligence, precision and stability of shotcrete quality control, reduces material waste and repair costs, and solves the problems in the above background technology.

[0006] In order to achieve the above purpose, the present application provides the following technical scheme: a tunnel shotcrete quality intelligent detection method based on multi-sensor fusion, comprising the following steps: S1, a spatial modeling process of the thermal radiation area of the shotcrete equipment is constructed, a thermal interference influence area prediction model is established according to the thermal conduction path and thermal radiation duration parameters of the spraying equipment under different operating conditions, and a thermal interference area prediction map is generated; S2, the thermal interference area prediction map is registered to the coordinate system of the infrared image in real time, a corresponding dynamic interference mask layer is constructed, and the high thermal interference area covered by the prediction map is shielded; S3, the thermal abnormal area is extracted from the infrared image after the interference area is removed, the thermal intensity gradient, boundary continuity coefficient and internal temperature change rate of each thermal abnormal area are calculated, and a structured thermal spot feature vector is generated; S4, based on the time sequence infrared image, the image morphological change trajectory and temperature decay trend of each thermal abnormal area are modeled, and the time sequence decay features of the thermal abnormal area are extracted; S5, the thermal spot feature vector and the time sequence decay features are input into a discriminative learning model, the misjudgment risk index of each thermal abnormal area is output, and a thermal spot confidence level label is generated according to the spatial structure consistency and time feature stability; S6, according to the thermal spot confidence level label, the parameter control operation of the shotcrete equipment is performed, the shotcrete parameter optimization adjustment is implemented for the thermal abnormal area with high confidence level, the feedback suppression and marking review operation is performed for the thermal abnormal area with low confidence level, and a closed-loop control system of infrared abnormality recognition and shotcrete parameter control is constructed.

[0007] Preferably, step S1 comprises: obtaining the heat source distribution parameters of the shotcrete equipment under different operating conditions; establishing a thermal conduction and thermal radiation model to simulate the diffusion path and intensity distribution of heat in space; generating a thermal interference area prediction map evolving with time, which is used to represent the spatial boundary of the concrete surface that may be affected by thermal interference; The thermal interference area prediction map is updated in real time when the device state changes for subsequent detection.

[0008] Preferably, step S2 comprises: establishing an imaging geometry model and a coordinate mapping relationship of the infrared imaging device; converting the boundary points of the thermal interference area prediction map into pixel-level positions of the infrared image; constructing a dynamic interference mask layer consistent with the size of the infrared image; embedding the dynamic interference mask layer into the infrared image analysis process to shield the high thermal interference area.

[0009] Preferably, step S3 comprises: extracting the thermal anomaly area using a dynamic threshold segmentation method; calculating the thermal intensity gradient of each thermal anomaly area; extracting the boundary continuity coefficient of the thermal anomaly area; calculating the internal temperature change rate of the thermal anomaly area and outputting the structured thermal spot feature vector.

[0010] Preferably, step S4 comprises: constructing a time series of infrared image frames and tracking the thermal anomaly area between frames; collecting the temperature values and spatial boundary shape parameters of the thermal anomaly area in each image frame; establishing a temperature decay model and a shape change model to fit the time evolution trend; constructing a time series decay feature vector from the temperature change slope, area change rate, and boundary complexity index.

[0011] Preferably, step S5 comprises: inputting the thermal spot structure feature vector and the time series decay feature vector into a discriminant learning model to output a misjudgment risk index; calculating the spatial structure consistency index and the time feature stability index of the thermal anomaly area; generating a confidence comprehensive score based on the misjudgment risk index, the spatial structure consistency index, and the time feature stability index using a weighted scoring mechanism; dividing the confidence comprehensive score into confidence level labels.

[0012] Preferably, after completing infrared thermal spot recognition and obtaining the confidence level labels of each thermal anomaly area, the discrete label values are converted into quantifiable response weight coefficients for driving fine control of subsequent spraying device parameters, thereby more comprehensively representing the processing priority of the current area. After obtaining the response weight coefficients, construct key parameter adjustment factors for controlling the spraying device, including spraying pressure adjustment amount, spraying angle adjustment amount, and spraying time adjustment amount. After obtaining the injection pressure adjustment amount, the injection angle adjustment amount, and the injection time adjustment amount, the operation instruction of the injection equipment is automatically updated according to the result, and whether the review mechanism is triggered is judged in combination with the weight threshold value, and the operation strategy is as follows: When the response weight coefficient is greater than the set response activation high threshold value, it is considered that the current area abnormality is highly reliable, and the following injection parameter update is immediately performed; When the response weight coefficient is less than the response shielding low threshold value, it is considered that the thermal anomaly area has no processing value, and the injection action is not performed, the parameter is set to 0, and the review process is entered: And the identification data of the current area is uploaded to the cloud review module for secondary artificial or AI rejudgment.

[0013] In the above technical solution, the technical effects and advantages provided by the present application are: The present application effectively shields the local thermal anomaly caused by the thermal radiation of the equipment by constructing the spatial modeling of the thermal radiation area of the injection equipment, real-time registering the coordinate system of the thermal interference area and the infrared image, and dynamically generating the interference mask layer, thereby avoiding the misidentification of the concrete surface quality defects. At the same time, by extracting the thermal spot feature vector and the time sequence attenuation feature, and combining the discriminant learning model to accurately evaluate the thermal anomaly area, a thermal spot confidence level label is generated, the injection parameter optimization adjustment of the high confidence thermal anomaly area is realized, and the feedback inhibition and review marking of the low confidence area are implemented, thereby constructing a closed-loop control system of infrared anomaly recognition and injection equipment parameter regulation and control. This scheme significantly improves the accuracy and intelligent level of the injection concrete quality control, reduces the material waste, over-spraying and subsequent repair cost caused by misjudgment, and ensures the construction quality and the long-term stability of the structure. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments or the prior art, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0015] Figure 1 The method flowchart of the present application is a tunnel injection concrete quality intelligent detection method based on multi-sensor fusion. DETAILED DESCRIPTION

[0016] Example implementations will now be described more fully with reference to the accompanying drawings. Example implementations may, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive gist to those skilled in the art.

[0017] The application provides a tunnel shotcrete quality intelligent detection method based on multi-sensor fusion, as shown in the following steps. Figure 1 S1, a spatial modeling process of a thermal radiation area of a shotcrete equipment is constructed, a thermal interference influence area prediction model updated with time is established according to thermal conduction paths and thermal radiation duration parameters of the shotcrete equipment in different running states, and a thermal interference area prediction map is generated. To improve the infrared detection accuracy in the shotcrete construction process and reduce the misjudgment risk caused by equipment thermal radiation, a spatial modeling process of a thermal radiation area of a shotcrete equipment is proposed, and a thermal interference influence area prediction model evolved with time is constructed. The specific implementation includes the following steps. The thermal source distribution parameters of the shotcrete equipment in multiple typical running states are obtained. The shotcrete equipment is divided into multiple running states according to the actual construction conditions, including a continuous spraying state, an intermittent waiting state and a spraying end state, and the temperature of key parts such as a nozzle, a mechanical arm, a motor shell and a hydraulic drive assembly is collected through a thermocouple sensor, an infrared temperature measuring device or a thermal imaging recording system. The device surface temperature distribution map corresponding to each running state is recorded continuously during the collection process, and a set of thermal source boundary condition data set of the equipment in different states is constructed in combination with the environmental temperature, humidity and ventilation condition of the equipment, thereby providing basic input for subsequent thermal propagation modeling. Based on the thermal conduction and thermal radiation theoretical model, the diffusion path of heat from the surface of the equipment to the surrounding shotcrete area is modeled. A thermal diffusion simulation model under the boundary condition of non-uniform structure is established by using a three-dimensional thermal conduction differential equation combined with a surface emissivity, an equipment shape structure and a relative spatial position relationship between the equipment and the concrete wall. The model introduces a radiation source intensity attenuation function and a spatial directionality coefficient to realize dynamic evaluation of the diffusion intensity of the equipment heat at different distances and angle directions. At the same time, to enhance the time dimension performance of the model, a time discretization method is used to construct a temperature change sequence in the thermal diffusion process, thereby realizing the predictability simulation of the thermal interference area evolved with time.

[0018]

[0019] ​Based on the above thermal diffusion modeling results, the thermal interference influence area prediction map of the spraying equipment under each operating state and its combination state is generated. The prediction map is expressed in the form of spatial thermal intensity distribution map, showing the area boundary and thermal interference level of the concrete surface that may be affected by equipment radiation at each time. To enhance the practical adaptability of the prediction map, a field calibration mechanism can be introduced to verify the accuracy of the simulation boundary in combination with the initial thermal image, and if there is deviation, the model parameters are automatically corrected. The prediction map is stored in the infrared image analysis system in the form of two-dimensional or three-dimensional layers as basic data for subsequent infrared image mask construction.

[0020] The thermal interference area prediction map is integrated with the infrared detection system, and a dynamic updating mechanism of the thermal interference area is established. During the spraying construction process, whenever the equipment operating state is switched or the position is moved, the system will call the thermal model under the corresponding state to recalculate the thermal interference area boundary at the current time, realizing real-time updating of the prediction map. In order to ensure the calculation efficiency of the system, the thermal model can be pre-fitted by finite element offline simulation or neural network regression model, so as to realize the rapid output of thermal interference prediction. This mechanism can ensure that the infrared detection system always operates on the basis of the latest thermal interference prediction map, effectively avoiding the interference of equipment thermal radiation on defect recognition accuracy, and thus improving the intelligence and reliability of the entire sprayed concrete quality detection system.

[0021] In summary, the thermal interference modeling process of the spraying equipment proposed in the present application fully considers the thermal source characteristics and dynamic propagation characteristics under the variable working conditions of the construction site, combines thermal physical modeling and sensor data fusion, and constructs a thermal interference prediction map with time updating capability, providing key support for subsequent infrared image processing, realizing precise shielding control of equipment thermal influence, and having significant technical creativity and engineering practicability.

[0022] S2, real-time registration of the thermal interference area prediction map to the coordinate system of the infrared image, construction of the corresponding dynamic interference mask layer for shielding the high thermal interference area covered by the prediction map; In order to effectively eliminate the infrared image misjudgment problem caused by equipment thermal radiation during sprayed concrete construction, after constructing the thermal interference area prediction map, a method of accurately registering the prediction map to the infrared image coordinate system and constructing a dynamic interference mask layer based on the registration result is further proposed to effectively shield the high thermal interference area during infrared image analysis. The method includes the following four steps: An imaging geometry model of the infrared imaging device and a coordinate conversion relationship are established. Specifically, the mapping relationship between the infrared image coordinate system and the three-dimensional coordinate system of the tunnel space is determined according to the installation position, field of view, focal length, sensor array size and image resolution of the infrared sensor. Through a calibration experiment, pixel coordinates of a plurality of known spatial position points in the infrared image are obtained, a perspective projection matrix is used to construct a space-image mapping function, and accurate mapping of the space thermal interference prediction map in the infrared image coordinate is realized.

[0023] A coordinate registration operation is performed on the thermal interference area prediction map, and the boundary points representing the thermal diffusion area in the prediction map are converted into pixel-level position information of the infrared image according to the coordinate mapping relationship. In order to improve the registration accuracy, an image geometry correction algorithm can be introduced to correct the image offset caused by sensor distortion or device vibration. In the registration process, it is necessary to ensure that the thermal interference area has a complete boundary closed structure in the infrared image, and an identifiable mask boundary is formed in the image coordinate, so as to avoid problems such as image occlusion discontinuity and region matching error.

[0024] A dynamic interference mask layer consistent with the size of the infrared image is constructed. The mask layer is a two-dimensional matrix corresponding to the pixels of the infrared image, and the pixel positions in the mask area are set by the registered thermal interference area. In order to realize dynamic updating of the mask, the present application introduces a real-time state monitoring mechanism, which switches the corresponding thermal interference prediction map in real time according to the state change of the spraying device (such as stopping, moving, spraying starting), and re-performs mapping and mask updating operation, to ensure that each frame of infrared image uses the current most matched interference mask, and to ensure the timeliness and accuracy of interference shielding.

[0025] The generated dynamic interference mask layer is embedded into the analysis process of the infrared image processing system, to realize automatic shielding of the high-thermal interference area in the thermal image. In the subsequent image analysis processes such as thermal spot identification, boundary extraction and feature calculation, the system sets the mask area pixels as invalid analysis area, which not only avoids misjudgment under the influence of device thermal radiation, but also preserves the information integrity of the real thermal anomaly area in the image. At the same time, in order to improve the environmental adaptability of the system, the present application also introduces a multi-source synchronization mechanism to synchronize the thermal interference prediction map generation module and the infrared image acquisition module at the frame level, to ensure that the mask and the image are one-to-one corresponding in the time axis, thereby constructing an infrared image interference shielding mechanism with spatial matching accuracy and time synchronization capability.

[0026] The embodiment provides a method for accurately registering a thermal interference area prediction map to an infrared image coordinate system and constructing a dynamic mask layer, which not only realizes accurate area isolation of thermal radiation interference of a spraying device, but also guarantees the recognition stability and robustness of an infrared detection system in a complex construction environment through dynamic updating and a multi-source synchronization mechanism, provides a reliable interference suppression capability for a sprayed concrete quality intelligent detection system, and has high technical advancement and practical engineering application value.

[0027] S3, extracting a thermal anomaly area in the infrared image after the interference area is removed, calculating a thermal intensity gradient, a boundary continuity coefficient and an internal temperature change rate of each thermal anomaly area, and generating a structured thermal spot feature vector; In order to realize accurate identification and quantitative analysis of the thermal anomaly area in the infrared image of sprayed concrete, after the thermal interference area is shielded, a thermal spot structured description method based on image feature calculation is further proposed, which aims to provide high-dimensional feature input for subsequent intelligent classification and risk discrimination. The method performs spatial segmentation and multi-dimensional feature quantization on the thermal anomaly area in the infrared image through a series of image processing and parameter extraction operations, and specifically includes the following steps: Based on the infrared image after the interference area is removed, an initial segmentation operation of the thermal anomaly area is performed. In this step, a thermal intensity dynamic threshold segmentation method is adopted, the local statistical adaptive threshold is set by analyzing the temperature distribution histogram of the image, and the thermal pixel set with temperature significantly higher than the surrounding background area is screened out. Then, morphological closing operation is used to repair the boundary and fill the holes of the initial thermal spot area, so that the segmented area has good boundary closure and region connectivity. In order to avoid the interference of isolated pixels, a minimum thermal spot area threshold can be set to remove noise points, and finally a plurality of thermal anomaly candidate regions in the image are obtained.

[0028] The thermal intensity gradient of each extracted thermal anomaly area is calculated. The thermal intensity gradient represents the descending rate of the temperature of the area from the center to the boundary, and is used to describe the internal temperature distribution uniformity and local concentration of the thermal spot. Specifically, a radial scanning path is constructed in each thermal anomaly area, the first derivative of the pixel temperature value on the path is calculated, and the average gradient amplitude and the distance proportion of the maximum gradient point located at the boundary are counted. This index can effectively reflect whether the thermal spot is generated by the internal energy source, so as to assist in distinguishing the thermal spot caused by the real structure defect from the false high-temperature area caused by thermal diffusion interference.

[0029] The boundary continuity coefficient of the extracted thermal anomaly region is extracted. The parameter is used to measure the shape regularity of the thermal spot edge and the boundary smoothness. The boundary tracking algorithm is used to extract the thermal spot contour, calculate the boundary curvature rate of change and the continuous distribution of the boundary gradient direction, and then the frequency spectrum features are extracted by the Fourier descriptor to evaluate the contour complexity. The higher the boundary continuity coefficient is, the more regular the thermal spot shape is, which usually corresponds to the real physical defects; while irregular edges are often found in thermal interference regions or local evaporation anomaly regions, thus having strong discriminability.

[0030] The internal temperature change rate of the thermal anomaly region is calculated, and a structured thermal spot feature vector is constructed. The temperature change rate refers to the consistency of the temperature distribution variance and the gradient direction of each pixel point in the region, which is used to measure the stability and physical consistency of the thermal spot internal temperature field. By establishing the pixel temperature matrix of the thermal anomaly region, the average temperature difference, standard deviation, and direction gradient consistency index are calculated, and together with the aforementioned thermal intensity gradient and boundary continuity coefficient, a structured feature description vector of the thermal spot is constructed. The vector is packaged in a unified format for output, providing input basis for the subsequent classification model, ensuring that the classification model can make judgments based on multiple information, rather than relying on a single temperature value or thermal spot area, thereby significantly improving the misjudgment suppression ability and the reliability of the classification decision.

[0031] The present embodiment introduces three core indicators of gradient, boundary, and region consistency based on the extraction of spatial thermal anomalies, forming a high-dimensional and highly discriminative thermal spot vector representation method, which provides a solid data foundation for subsequent misjudgment risk assessment and thermal spot confidence discrimination based on machine learning, and has strong innovation and engineering feasibility.

[0032] S4, based on time series infrared images, modeling the image shape change trajectory and temperature decay trend of each thermal anomaly region, extracting the time series decay features of the thermal anomaly region; In order to further enhance the recognition ability of the infrared detection system for the thermal anomaly region and improve the misjudgment suppression effect, on the basis of completing the spatial feature extraction of the thermal anomaly region, a dynamic feature modeling method based on time series infrared images is proposed, which extracts the image shape evolution trajectory and temperature decay trend of the thermal spot region in consecutive frames of images, constructs the time series decay features, and distinguishes the real structure defect thermal spot from the false thermal spot generated by the device thermal diffusion interference. The method includes the following four implementation steps: The time series data structure of the infrared image sequence is constructed and the frame-level region tracking is implemented. By continuously acquiring multiple thermal images captured by the fixed infrared imaging equipment in the construction site within a unit time interval, a set of image frame sequences with time markers is constructed. For the thermal anomaly regions extracted in the initial frame, a region feature template matching method is used to track the pixel-level of the corresponding regions in the subsequent image frames. This method combines the position coordinates, contour shape and temperature distribution template of the thermal spot to realize cross-frame target consistency judgment, ensuring the numbering consistency and trajectory continuity of the same thermal anomaly region in multiple image frames, and providing reliable input data basis for subsequent time series analysis.

[0033] The temperature attributes and spatial boundary shape information of each thermal anomaly region at different time points are collected. In the image frames where the region matching has been completed, the key feature parameters of each thermal anomaly region such as the maximum temperature value, average temperature value, boundary area, perimeter, center position, etc. are extracted, and a multi-dimensional time series feature curve based on time stamp is constructed. For example, the decay curve of temperature with time, the expansion or contraction curve of area with time, etc. To improve the analysis accuracy, boundary shape complexity indicators such as curvature variance or boundary density can also be introduced to capture the subtle trends of the region shape changes. These time series data collectively constitute the dynamic attribute description of the thermal anomaly region, which is used for subsequent modeling analysis.

[0034] Based on the collected time series attribute data, the temperature decay model and the shape change model of the thermal anomaly region are established. In terms of temperature decay modeling, an exponential decay function or a piecewise linear regression model is used to fit the temperature change trend of the thermal spot to characterize its behavior characteristics from high temperature to ambient temperature. Real defect thermal spots usually decay slowly or relatively stably, while device thermal radiation interference thermal spots show a clear rapid decay trend. In terms of shape change modeling, a time window moving average strategy is used to fit and smooth the area and boundary curvature changes of the thermal spot region to extract its spatial evolution law. By integrating the above two models, a unified time series decay feature vector can be generated, which fully reflects the behavior pattern of the thermal anomaly region in the time dimension.

[0035] The constructed time sequence attenuation feature vector is used in subsequent classification model or discrimination system as a key judgment basis for identifying the nature of thermal anomaly. In order to enhance the feature expression ability, the embodiment further proposes a time sequence multi-dimensional feature fusion strategy, and the temperature change curve slope, fitting residual, area change rate, boundary complexity index and other feature quantities are normalized and combined into a standard vector format, which is uniformly input into the classification module. The vector can be used jointly with the structured thermal spot space feature vector to construct a space-time multi-dimensional feature system, which significantly improves the recognition ability of high-similarity thermal spots. Especially in the case of high overlap between device thermal interference and real air hole thermal spots in image representation, the introduction of time sequence attenuation features can significantly enhance the model's ability to judge the "thermal source origin" and reduce the problem of control command mis-triggering caused by instantaneous high temperature misidentification.

[0036] The embodiment introduces a time sequence dynamic modeling mechanism of thermal anomaly area in the infrared image analysis process, fully utilizes the evolution law of thermal spot in time dimension, realizes the technical breakthrough from static image recognition to dynamic behavior recognition, provides key support for constructing a high-robustness thermal anomaly intelligent detection system, and has outstanding technical creativity and practical engineering value.

[0037] S5, input the thermal spot feature vector and the time sequence attenuation feature into the discrimination learning model, output the misjudgment risk index of each thermal anomaly area, and generate the thermal spot confidence level label according to the spatial structure consistency and the time feature stability; In order to realize the reliability evaluation of the thermal anomaly area recognition result in the infrared image, and further reduce the misjudgment probability caused by device thermal interference, after completing the thermal spot space feature extraction and time sequence modeling, a thermal anomaly misjudgment risk evaluation method based on a discrimination learning model is proposed, a unified input feature vector is constructed, spatial structure consistency and time feature stability are introduced as auxiliary criteria, misjudgment risk index of each thermal anomaly area is output, and corresponding confidence level label is generated, which provides reliable classification basis for subsequent spraying equipment control system. The method specifically includes the following four steps: Constructing a multi-dimensional input feature vector of thermal anomaly area. The spatial structure features (including thermal intensity gradient, boundary continuity coefficient, internal temperature change rate) extracted in the previous steps are fused with the time sequence dynamic features (including temperature decay rate, area change trend, boundary shape evolution index), and a high-dimensional feature vector is organized in a unified format. In order to enhance the consistency and learnability of the model input, the features of each dimension are normalized, and the outliers are corrected by truncation or standard deviation constraint strategy, so as to construct a structured data set suitable for model input. Each thermal anomaly area corresponds to a unique input feature vector, which provides data basis for the classification model.

[0038] A misjudgment risk assessment model based on discriminative learning algorithm is constructed and trained. The model can be implemented by support vector machine (SVM), random forest, gradient boosting tree or lightweight neural network structure, and the training data is composed of infrared image manually annotated samples collected in the true manufacturing environment, covering two types of instances of real structural defect hot spots and equipment thermal interference hot spots. In the model training stage, the objective function is to minimize the misjudgment rate and maximize the recognition accuracy, cross-validation is used to evaluate the generalization ability, and feature importance analysis is performed on different feature combination schemes to ensure that the selected feature dimension has strong distinguishing performance and low redundancy. After the model training is completed, it has the ability to map the hot spot feature input to the misjudgment risk index output.

[0039] The trained discriminative model is used to predict the risk of the thermal anomaly area to be analyzed. The constructed high-dimensional feature vector is input into the discriminative learning model one by one, and the model classifies the type of the thermal anomaly area according to the internal decision boundary and outputs a continuous misjudgment risk index. The index can be set as a real value between 0 and 1, and the higher the value, the more likely it is that the thermal anomaly is a misjudgment area caused by equipment thermal diffusion; on the contrary, the lower the value, the more likely it is that it is a real structural defect. To enhance the interpretability of the risk index, the model output can be supplemented with feature contribution scores to reveal the feature dimensions that have the greatest impact on the current judgment, facilitating subsequent model optimization and expert intervention.

[0040] According to the misjudgment risk index output by the discriminative model, combined with the spatial structure consistency and time feature stability of the thermal anomaly area, a confidence level label is generated. The spatial structure consistency is evaluated by calculating the region boundary change rate and internal gradient consistency coefficient; the time feature stability is quantified by the fitting residual and time consistency of the temperature change curve. The above two-dimensional indicators and the risk index are jointly calculated to generate a confidence score using a weighted scoring mechanism. The score value is divided into multiple levels within a pre-set threshold interval, such as high confidence, medium confidence, and low confidence. The label result serves as the basis for subsequent device parameter regulation, and high-confidence hot spots trigger jet reinforcement operations, while low-confidence hot spots are automatically shielded or marked for review, thereby realizing intelligent feedback control loop based on data-driven.

[0041] This embodiment significantly improves the accuracy of classifying thermal anomaly areas by constructing a high-dimensional input vector that integrates spatial features and time sequence features, and introducing discriminative learning algorithms and confidence label mechanisms. It also provides accurate data support for subsequent automatic jet parameter regulation, effectively reduces the risk of misidentification caused by equipment thermal interference, and has obvious technical innovation and practical engineering application value.

[0042] S6. Perform parameter control operation of the shotcrete equipment according to the thermal spot confidence level label, implement shot parameter optimization adjustment on the thermal anomaly area with high confidence level, and perform feedback suppression and review operation on the thermal anomaly area with low confidence level, to build a closed-loop control system of infrared anomaly recognition and shotcrete equipment parameter regulation; The role of this step is to effectively link the thermal spot confidence level label output by the infrared image intelligent recognition system with the parameter control module of the shotcrete equipment, to build a closed-loop control system with autonomous perception, intelligent judgment and dynamic response capability, and to realize real-time optimization management of the shotcrete quality. Specifically, when the system determines that a certain area has high confidence level of thermal anomaly, it means that it is very likely to represent a real structural defect, such as hollow, unhydrated or honeycomb-like non-dense area. At this time, the system automatically activates the parameter regulation function of the shotcrete equipment, such as adjusting the shot pressure, material ratio, angle and duration, to accurately repair the defect and improve the shot quality and layer uniformity. On the contrary, for the thermal anomaly area with low confidence level, the system identifies it as a device thermal interference or misjudgment area, and immediately triggers the feedback suppression mechanism to avoid blind secondary shot, while marking the area and uploading it to the task scheduling system for manual or system review. Through this step, the entire system realizes a complete closed loop from infrared image recognition, intelligent classification to automatic feedback control, significantly improving the intelligent level and resource utilization efficiency of shot quality control on the construction site, reducing the risk of misoperation and material waste, and is a key link to realize fine and intelligent construction management.

[0043] After completing the infrared thermal spot recognition and obtaining the confidence level label of each thermal anomaly area, the discrete label value is converted into a quantifiable response weight coefficient for driving the fine control of the subsequent shotcrete equipment parameters. The construction of the response weight not only reflects the importance of the thermal spot, but also considers the area size of the thermal spot, so as to more comprehensively represent the processing priority of the current area. The calculation expression is as follows , wherein, is the response weight coefficient, representing the response priority and control sensitivity of the th thermal anomaly area in the current frame, with a value range of floating point numbers between 0 and 1. The larger the value, the more likely it is that the area is a structural defect, requiring more aggressive shot response. is the thermal spot confidence level label, representing the confidence level of the th thermal anomaly area, with 3 representing high confidence (very likely to be a real defect), 2 representing medium confidence (with some suspected defects), and 1 representing low confidence (suspected to be interference or non-defect). is the maximum value of the confidence level of all thermal spots in the current frame. is the normalized thermal anomaly area, representing the the ratio of the pixel area of the hot anomaly region to the maximum hot spot area, is a confidence influence weight coefficient, controlling the influence proportion of the confidence level on the response weight , with a value range of 0.6-0.8, if increased , the system will rely more on the image recognition result for response, is an area influence weight coefficient, used to control the influence intensity of the hot spot area on the response weight, with a value range of 0.2-0.4, increasing the control right of the hot spot physical size on the response mechanism, preventing the system from only relying on high-confidence small hot spots for control and ignoring possible macro defects; The role of this step is to realize the quantitative evaluation of the hot spot processing priority, providing the basic driving factor for the subsequent adaptive calculation of the spraying parameters.

[0044] After obtaining the response weight coefficient , the key parameter adjustment factor for controlling the spraying equipment is constructed, including the spraying pressure adjustment amount, the spraying angle adjustment amount, and the spraying time adjustment amount, each adjustment factor being calculated based on the specific physical and thermodynamic characteristics of the hot spot, combined with the response weight, and the calculation expression is as follows: , wherein is the spraying pressure adjustment amount of the i-th hot anomaly region, used to determine whether to increase the spraying pressure in the current spraying operation and the amplitude of the increase, is the maximum adjustment coefficient of the spraying pressure, which is the maximum pressure adjustment amplitude allowed, used to limit the upper limit of regulation and control, and avoid excessive correction, is the temperature fluctuation amplitude of the i-th hot anomaly region, representing the difference between the maximum value and the average value of the temperature in the hot anomaly region, reflecting whether the thermal field is concentrated, and a large fluctuation indicates that there is a central high-temperature hot spot in the region, which requires the increase of the pressure and the spraying time to ensure the repair depth, is the adjustment coefficient of the temperature fluctuation amplitude , controlling the sensitivity of the temperature fluctuation to the pressure, is the spraying angle adjustment amount of the i-th hot anomaly region, representing the angle size of the i-th hot anomaly region that needs to be adjusted in the direction of the nozzle during construction, is the maximum adjustment coefficient of the spraying angle, which is the maximum pressure adjustment amplitude allowed by the system, used to limit the upper limit of regulation and control, and avoid excessive correction, is the boundary density index of the i-th hot anomaly region, representing the i-th hot anomaly region ​​​​​The number of boundary pixels contained in a unit length of the contour line of a thermal anomaly area, which reflects the complexity of the thermal spot contour; a large value indicates irregular boundaries, which may be misjudgment areas, affecting the angle correction decision, is the adjustment coefficient of the boundary density index, which controls the influence of boundary density on angle correction, is the injection time adjustment amount of the first thermal anomaly area, which reflects the overall detection reliability and treatment priority of the first thermal anomaly area, and determines the intensity of parameter adjustment in subsequent injection control. It is the core weight of the control logic, is the maximum injection time adjustment coefficient, which defines the maximum time extension allowed under extreme conditions to prevent over-injection or insufficient injection, is the time stability index of the first thermal anomaly area, which represents the rate of change of the temperature of the thermal spot with time (taking the negative slope). The smaller the value, the more stable it is. It is used to evaluate whether the thermal anomaly area has the temperature stability that a real structural defect should have, so as to decide whether to extend the injection time, is the adjustment coefficient of the time stability index , which controls the adjustment range of the time stability on the injection duration; The purpose of this step is to combine the response weight with the characteristics of the thermal spot to generate multi-dimensional control parameters, so that each injection control action has precise adjustment capability driven by data.

[0045] After obtaining , and , the operation instructions of the injection equipment are automatically updated according to the results, and whether to trigger the review mechanism is judged in combination with the weight threshold value. The operation strategy is as follows: When the response weight coefficient is greater than the set response activation high threshold (such as 0.7), it indicates that the current area anomaly is highly reliable, and the following injection parameter update is immediately executed: , wherein is the final injection pressure, which represents the actual injection pressure value that the equipment will use in the thermal anomaly area, is the basic injection pressure, which represents the conventional injection pressure value of the injection concrete equipment under the condition of no thermal anomaly intervention. It is usually set by the equipment or set by the construction personnel according to the material strength and construction requirements, is the final injection angle, which represents the adjusted nozzle direction of the equipment in the thermal anomaly area, is the basic injection angle, which represents the standard working angle between the nozzle and the concrete sprayed surface. It is usually set to normal injection to ensure uniform coverage of the concrete, is the final spraying duration, representing the specific spraying duration that the device will perform in the area, is the basic spraying time, representing the standard spraying duration of the spraying device per unit area in the normal working state, used to ensure the spraying thickness, uniformity and structural strength; If is less than the response shielding low threshold (0.3, for example), the thermal anomaly area is considered to have no treatment value, and no spraying action is performed. The parameter is set to 0, and the review process is entered: The identification data of the current area is uploaded to the cloud review module for secondary manual or AI review.

[0046] This step applies the parameter adjustment result to the device control layer, sets the response suppression and recording mechanism for low confidence areas, builds the risk tolerance and error correction capability of the system, and truly realizes the adaptive closed-loop response mechanism between detection, decision and control.

[0047] Through the above-mentioned intelligent detection method for tunnel sprayed concrete quality based on multi-sensor fusion, the misjudgment problem caused by device thermal radiation interference in the traditional infrared detection system can be effectively solved. The method builds a spatial modeling of the thermal radiation area of the spraying device, real-time registers the coordinate system of the thermal interference area and the infrared image, and dynamically generates an interference mask layer, effectively shielding the local thermal anomaly caused by device thermal radiation, thereby avoiding misidentification of concrete surface quality defects. At the same time, by extracting the thermal spot feature vector and time sequence attenuation feature, and combining the discriminant learning model to accurately evaluate the thermal anomaly area, a thermal spot confidence level label is generated, the spraying parameter optimization adjustment of the high confidence thermal anomaly area is realized, and the feedback suppression and review marking of the low confidence area are implemented, thereby building an adaptive and accurate closed-loop control system of infrared anomaly identification and spraying device parameter regulation. This scheme significantly improves the accuracy and intelligent level of sprayed concrete quality control, reduces material waste, over-spraying and subsequent repair costs caused by misjudgment, and ensures the long-term stability of construction quality and structure.

[0048] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula closest to the actual situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0049] It is apparent that for the person skilled in the art, many modifications and changes can be suggested without departing from the scope of the application, and it is intended to encompass these modifications and changes as fall within the scope of the application.

[0050] It should be noted that, in the present document, relational terms are used solely to estab- lish a correspondence between particular entities and other entities, not necessarily a direct orocedural relationship between these entities. The terms "comprises", "comprising", or any other variational phrasing thereof are intended to encompass non-exclusive in- clusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. The terms "comprises", "comprising", or any other variational phrasing thereof are intended to encompass non-exclusive in- clusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus.

[0051] It should be understood that, in various embodiments of the present application, the sequence of the above processes does not mean the execution order, and the execution order of the processes should be determined according to their functions and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0052] Those skilled in the art can clearly understand that the units and algorithm steps of the examples described in combination with the embodiments disclosed in the present document can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0053] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0054] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, i.e. they can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments of the present application.

[0055] In addition, each functional unit in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit.

[0056] The above merely describes some exemplary embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

[0057] The above only describes some exemplary embodiments of the present application by way of illustration, and it is needless to say that the described embodiments can be modified in various ways without departing from the spirit and scope of the present application for those skilled in the art. Therefore, the above drawings and descriptions are illustrative in nature and should not be understood as limiting the protection scope of the claims of the present application.

Claims

1. A multi-sensor fusion-based intelligent detection method for tunnel shotcrete quality, characterized in that, The method comprises the following steps: S1, constructing a spatial modeling process of a thermal radiation area of a shotcrete device, establishing a time-updated thermal interference influence area prediction model according to thermal conduction paths and thermal radiation duration parameters of the shotcrete device in different operating states, and generating a thermal interference area prediction map; S2, real-time registration of the thermal interference area prediction map to a coordinate system of an infrared image, construction of a corresponding dynamic interference mask layer for shielding a high thermal interference area covered by the prediction map; S3, extraction of a thermal anomaly area in the infrared image after interference area elimination, calculation of thermal intensity gradients, boundary continuity coefficients and internal temperature change rates of each thermal anomaly area, and generation of a structured thermal spot feature vector; S4, modeling of image morphological change trajectories and temperature decay trends of each thermal anomaly area based on time series infrared images, and extraction of time series decay features of the thermal anomaly area; S5, input of the thermal spot feature vector and the time series decay feature into a discriminative learning model, output of a misjudgment risk index of each thermal anomaly area, and generation of a thermal spot confidence level label according to spatial structure consistency and time feature stability; S6, execution of parameter control operations of the shotcrete device according to the thermal spot confidence level label, implementation of shotcrete parameter optimization adjustment on thermal anomaly areas with high confidence level, and execution of feedback suppression and marked review operations on thermal anomaly areas with low confidence level, and construction of a closed-loop control system of infrared anomaly recognition and shotcrete device parameter regulation. 2.The multi-sensor fusion based tunnel sprayed concrete quality intelligent detection method according to claim 1, characterized in that, Step S1 comprises: acquiring thermal source distribution parameters of the shotcrete device in different operating states; establishing a thermal conduction and thermal radiation model to simulate the diffusion path and intensity distribution of heat in space; generating a time-evolving thermal interference area prediction map for representing the spatial boundary of the concrete surface that may be subjected to thermal interference; updating the thermal interference area prediction map in real time when the device state changes for subsequent detection. 3.The multi-sensor fusion based tunnel sprayed concrete quality intelligent detection method according to claim 1, characterized in that, Step S2 comprises: establishing an imaging geometry model and coordinate mapping relationship of the infrared imaging device; converting boundary points of the thermal interference area prediction map to pixel-level positions of the infrared image; constructing a dynamic interference mask layer consistent with the size of the infrared image; embedding the dynamic interference mask layer into the infrared image analysis process to shield the high thermal interference area.

4. The multi-sensor fusion-based intelligent detection method for tunnel sprayed concrete quality according to claim 1, characterized in that, Step S3 comprises: extracting the thermal anomaly area by using a dynamic threshold segmentation method; calculating the thermal intensity gradient of each thermal anomaly area; extracting the boundary continuity coefficient of the thermal anomaly area; calculating the internal temperature change rate of the thermal anomaly area and outputting a structured thermal spot feature vector.

5. The multi-sensor fusion-based intelligent detection method for tunnel sprayed concrete quality according to claim 1, characterized in that, Step S4 comprises: constructing time series infrared image frames and tracking the thermal anomaly area between frames; collecting temperature values and spatial boundary morphological parameters of the thermal anomaly area in each image frame; establishing a temperature decay model and a morphological change model to fit the time evolution trend; constructing a time series decay feature vector with the temperature change slope, area change rate and boundary complexity index. 6.The multi-sensor fusion based intelligent detection method for tunnel sprayed concrete quality according to claim 1, characterized in that, Step S5 comprises: inputting the thermal spot structure feature vector and the time series decay feature vector into a discriminative learning model to output a misjudgment risk index; calculating the spatial structure consistency index and the time feature stability index of the thermal anomaly area; According to the misjudgment risk index, the spatial structure consistency index and the time characteristic stability index, a weighted scoring mechanism is used to generate a confidence comprehensive score; According to the confidence comprehensive score, a confidence level label is divided.

7. The multi-sensor fusion-based intelligent detection method for tunnel sprayed concrete quality according to claim 1, characterized in that, After completing the infrared hot spot recognition and obtaining the confidence level label of each thermal anomaly area, the discrete label value is converted into a quantifiable response weight coefficient for driving the fine control of the subsequent spraying equipment parameters, so as to more comprehensively represent the processing priority of the current area; After obtaining the response weight coefficient, a key parameter adjustment factor for controlling the spraying equipment is constructed, including a spraying pressure adjustment amount, a spraying angle adjustment amount and a spraying time adjustment amount; After obtaining the spraying pressure adjustment amount, the spraying angle adjustment amount and the spraying time adjustment amount, the operation instruction of the spraying equipment is automatically updated according to the result, and whether the review mechanism is triggered is judged in combination with the weight threshold value, and the operation strategy is as follows: When the response weight coefficient is greater than the set response activation high threshold value, it indicates that the current area abnormality is highly reliable, and the following spraying parameter update is immediately executed; If the response weight coefficient is less than the response shielding low threshold value, it is considered that the thermal anomaly area has no processing value, and no spraying action is performed, the parameter is set to 0, and the review process is entered: And the recognition data of the current area is uploaded to the cloud review module for secondary artificial or AI review.

Images