High formwork support system monitoring method, system and device

Abstract

The application discloses a high and large template support system monitoring method, system and equipment, belongs to the field of building intelligent monitoring, and the method comprises the steps of: embedding a three-way strain sensor in the concrete of the bottom plate directly below each vertical rod to measure three orthogonal direction strain components; establishing a finite element model to extract three-way normal stress at the corresponding position to form a theoretical stress field; after erecting the support system and pouring the roof concrete, collecting strain data and converting them into a measured stress field; calculating the point-by-point deviation of the measured and theoretical stress fields to generate a three-channel stress deviation field; for each vertical rod, extracting a local window three-channel stress deviation matrix centered on the vertical rod, inputting a pre-trained convolutional neural network model, and outputting the disease type of the vertical rod. The application converts the bottom plate stress monitoring into an image recognition problem by using the full-scaffold matrix regular layout, and realizes global intelligent diagnosis of all hidden diseases of the vertical rods.

Description

Technical Field

[0001] This invention relates to the field of intelligent building monitoring. More specifically, this invention relates to methods, systems, and equipment for monitoring tall formwork support systems. Background Technology

[0002] High-rise formwork support systems are temporary load-bearing structures used in building construction for the construction of horizontal components such as cast-in-place concrete slabs and large-span beams. Full-span scaffolding is widely used due to its high load-bearing capacity and good overall integrity. Full-span scaffolding is typically assembled from a large number of steel pipe uprights, horizontal bars, and scissor braces at designed intervals. The uprights are arranged in a matrix with equal intervals in the plane, with longitudinal and transverse spacing generally ranging from 0.6 to 1.2 meters. The entire support system has a large number of densely distributed members. During concrete pouring, the load borne by the uprights is transferred to the underlying slab through the base. Even minor abnormalities in the uprights, such as partial bottom suspension, top support misalignment, pole tilting, or failure of horizontal bar connections, can alter the force transmission path, causing localized stress concentration and redistribution. In severe cases, this can lead to instability or even complete collapse of the support system.

[0003] Existing full-span scaffolding monitoring methods largely rely on manual inspections and localized instrument monitoring. Manual inspections are limited by dense scaffolding, obstructed views, and personnel experience, making it difficult to detect early-stage defects such as hidden gaps at the base of uprights or detachment of the top scaffolding. Instrument monitoring typically involves deploying axial force gauges, displacement sensors, or inclinometers at only a few representative uprights, inferring the system's safety status through changes in axial force or displacement at individual points. This approach cannot cover the vast number of uprights. For the overall support system with complex load transfer paths and diverse defect locations and forms, localized monitoring information cannot comprehensively reflect the true working condition of each upright, nor can it systematically diagnose the specific anomalies occurring on each upright. Simply relying on threshold alarms cannot achieve accurate identification of defect types and locations.

[0004] The uprights of a full-span scaffold are arranged in a regular, matrix-like, equally spaced pattern in a plane. This geometric feature creates a natural spatial discrete grid in the base plate area directly beneath each upright. When the uprights transfer the upper load to the base plate concrete, vertical and two horizontal normal stress components are generated within the base plate. Any change in the state of the uprights will cause characteristic deviations in these three stress components locally at their bottom. This need to obtain three independent stress components at each grid node and perform spatial analysis is highly similar in data form to pixel arrays and RGB three-channel structures in digital images. In recent years, image recognition technology based on convolutional neural networks has demonstrated powerful spatial feature extraction capabilities in areas such as defect detection and pattern classification, and can automatically learn complex local patterns from multi-channel gridded data. Inspired by this, if the three-dimensional stress deviation field at each measuring point of the support system base plate is analogized to a three-channel image, a deep learning model can be used to identify the spatial patterns of stress anomalies as a whole, thereby providing a monitoring method, system, and equipment for tall formwork support systems. Summary of the Invention

[0005] One objective of this invention is to provide a monitoring method, system, and device for tall formwork support systems. For full-span scaffolding with a regular matrix layout, a three-dimensional strain sensor is pre-embedded in the base plate below the uprights to form a three-channel stress deviation field. By using a convolutional neural network to learn stress anomaly patterns, a global intelligent diagnosis of hidden defects such as suspension and tilting at the bottom of each upright can be achieved, breaking through the limitations of local monitoring and comprehensively improving the accuracy and efficiency of construction safety monitoring.

[0006] To achieve these objectives and other advantages according to the invention, according to one aspect of the invention, the invention provides a monitoring method for a tall formwork support system, comprising the following steps: S1. Based on the planar arrangement of the support system poles, a strain sensor is pre-embedded in the base plate concrete directly below each pole. The strain sensor is used to measure the strain components in three orthogonal directions within the concrete. S2. Establish a finite element model including the base plate, support system and top load, simulate the stress distribution in the base plate under load, and extract the normal stress in three orthogonal directions at the corresponding positions of the strain sensor to form the theoretical stress field of the base plate. S3. Erect the support system and complete the concrete pouring of the top slab. Collect the strain component data of all strain sensors and convert the strain component data into the measured stress field of the bottom slab. S4. Calculate the point-by-point deviation between the measured stress field and the theoretical stress field, and generate a three-channel stress deviation field composed of stress deviation values ​​in three orthogonal directions. S5. For each pole, using a pre-trained convolutional neural network model, based on the three-channel stress deviation field data of each measuring point in the local area centered on the pole, the type of damage to the pole is diagnosed, and the diagnosis result is output. S6. Summarize the diagnostic results of all poles, mark the locations of abnormal poles, and generate reinforcement plans.

[0007] Preferably, 2. the strain sensor consists of a vertical steel bar and three orthogonal strain gauges attached thereon, which measure the strain in the vertical, first horizontal and second horizontal directions respectively; the strain sensor is vertically fixed in the concrete base plate directly below the base of each pole, and the horizontal measurement direction of all strain sensors is the same.

[0008] Preferably, test blocks cured under the same conditions are prepared simultaneously when pouring the base slab concrete, and strain compensation sensors are embedded in the test blocks. After the base slab concrete reaches the design strength, all strain sensors and the strain compensation sensors are simultaneously set to zero. In subsequent strain data processing, the measured strain value of each strain sensor is subtracted from the free strain measured by the strain compensation sensor at the same time to deduct the influence of concrete shrinkage and temperature effect on stress calculation.

[0009] Preferably, step S5 includes: Centered on the current pole, a local window covering the pole and its adjacent poles is extracted from the three-channel stress deviation field. The stress deviation values ​​of each measuring point in the local window in the vertical, first horizontal and second horizontal directions are respectively used as the values ​​of the pole in the three independent channels. The measuring points are arranged according to their planar positions to form the three-channel stress deviation matrix of the local window. The three-channel stress deviation matrix is ​​input into a pre-trained convolutional neural network model; The spatial distribution characteristics of the three-dimensional stress deviation in the three-channel stress deviation matrix are extracted by the convolutional neural network model, and the defect type of the pole is output accordingly. The defect type includes at least one of the following: pole bottom suspension, top support eccentricity, pole tilting, and horizontal connection failure.

[0010] Preferably, the convolutional neural network model includes: The input layer is used to receive the three-channel stress deviation matrix; The first convolutional layer is used to simultaneously extract local spatial features from the three channels of the three-channel stress deviation matrix and output the first intermediate result. The second convolutional layer is used to further combine features from the first intermediate result and output the second intermediate result. A flattening layer is used to stretch the second intermediate result into a one-dimensional vector; A fully connected layer is used to integrate the one-dimensional vector to obtain a representation of the disease characteristics of the pole. The output layer is used to output the probability that the pole belongs to each type of disease based on the disease characteristics.

[0011] Preferably, the output layer further includes a regression output branch, which is used to output multiple disease severity parameters; the convolutional neural network model selects effective parameters corresponding to the disease type from the multiple disease severity parameters according to the classification results; when the value of the effective parameter exceeds a preset severity threshold, the disease of the pole is immediately treated on-site.

[0012] Preferably, the training process of the convolutional neural network model includes: Define a joint loss function, which is composed of a weighted sum of a classification loss term and a regression loss term. The classification loss term is used to measure the difference between the predicted value of the disease type and the true label, and the regression loss term is used to measure the difference between the predicted value of the disease severity parameter and the true label. The convolutional neural network model to be trained is trained using a pre-built training sample set. The network weights are iteratively updated by minimizing the joint loss function and using the backpropagation algorithm until the model converges.

[0013] Preferably, the construction of the training sample set includes: Training samples are generated by parametric finite element model. Different types of pole defects are set in the finite element model. After applying construction load, the triaxial stress deviation field of the area around the pole is extracted. A three-channel stress deviation matrix is ​​constructed and the corresponding defect type and defect severity parameters are labeled as tags. Each training sample contains a three-channel stress bias matrix and its corresponding label.

[0014] A monitoring system for tall formwork support systems, applied to the aforementioned monitoring methods for tall formwork support systems, including: The strain sensor array is embedded in the concrete base plate directly below each pole; The data acquisition device is electrically connected to the strain sensor array and is used to acquire strain data from each strain sensor during and after the pouring of the top slab concrete. A data processing device, communicatively connected to the acquisition device, the data processing device comprising: The stress field generation module is used to convert the collected strain data into a measured stress field and calculate the deviation from the pre-stored theoretical stress field point by point to generate a three-channel stress deviation field. The diagnostic module has a built-in pre-trained convolutional neural network model, which is used to output the type of damage to the pole based on the three-channel stress deviation field data of each measuring point in a local area centered on each pole. The early warning module is used to summarize the diagnostic results of all poles, mark the locations of abnormal poles, and generate reinforcement plans.

[0015] A computer device includes a memory and a processor, wherein the memory stores computer instructions, and the processor executes the computer instructions to perform the above-described high-mountain template support system monitoring method.

[0016] The present invention has at least the following beneficial effects: First, by shifting the monitoring object from the pole itself to the interaction interface between the pole and the base plate, and by pre-embedding a three-dimensional strain sensor in the base plate directly below each pole, this invention can capture the local stress redistribution characteristics of the base plate caused by hidden defects that are difficult to detect by direct monitoring methods, such as pole bottom suspension and top support eccentricity, thereby fundamentally improving the sensitivity and comprehensiveness of defect identification.

[0017] Secondly, by making full use of the natural geometric features of the regular matrix arrangement of the uprights of the full-span scaffolding, the present invention constructs the triaxial stress deviation of each measuring point into a three-channel deviation field with a spatial structure, which is equivalent to transforming the structural mechanics problem into an image recognition problem. This enables the convolutional neural network to efficiently extract the unique spatial stress deviation patterns of various defects, achieving comprehensive coverage of all uprights in the entire support system.

[0018] Third, this invention uses a pre-trained convolutional neural network model to automatically analyze the three-channel stress deviation field, which can simultaneously diagnose various types of defects such as pole bottom suspension, pole body tilting, and horizontal connection failure, and assess their severity. This breaks through the limitation of traditional methods that can only detect load-bearing capacity exceeding limits but cannot distinguish specific causes, and provides a reliable basis for precise reinforcement. Attached Figure Description

[0019] Figure 1 This is an overall flowchart of one technical solution of the present invention; Figure 2 This is a schematic diagram of a tall, full-span scaffold structure for formwork support in one technical solution of the present invention; Figure 3 This is a schematic diagram of a three-channel stress deviation field in one technical solution of the present invention; Figure 4 This is a schematic diagram of a convolutional neural network model in one of the technical solutions of the present invention. Detailed Implementation

[0020] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can implement it based on the description.

[0021] It should be understood that terms such as “having,” “comprising,” and “including” as used herein do not exclude the presence or addition of one or more other elements or combinations thereof.

[0022] It should be noted that, unless otherwise specified, the experimental methods described in the following embodiments are conventional methods, and the reagents and materials mentioned are commercially available. In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "setting" should be interpreted broadly. For example, they can refer to fixed connection or setting, detachable connection or setting, or integral connection or setting. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The terms "lateral," "longitudinal," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing this invention and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of this invention.

[0023] like Figures 1-4 As shown, this invention provides a monitoring method based on a tall template support system, comprising the following steps: S1. Based on the planar layout of the support system poles, a strain sensor is pre-embedded in the base slab concrete directly below each pole. The strain sensor is used to measure the strain components in three orthogonal directions within the concrete.

[0024] In practical implementation, the specific coordinates of all uprights on the plane are first determined according to the design scheme of the high-rise formwork support system. The uprights of the full-span support system are usually arranged in a matrix with equal spacing, and the longitudinal and transverse spacing are clearly specified in the construction drawings. Before pouring the base slab concrete, on the working surface after the base slab reinforcement is tied, lines are laid out on-site according to the upright layout drawing, marking the corresponding points on the base slab directly below the base of each upright. These points are the locations for embedding strain sensors. This one-to-one correspondence arrangement ensures that after the subsequent support system is erected, the load-bearing area at the bottom of each upright is exactly above a strain sensor, thus ensuring that the sensor can directly sense the stress changes within the base slab concrete when the upright transmits the load to the base slab.

[0025] Before installation, the strain sensors must undergo necessary waterproofing, insulation, and mechanical protection to adapt to the construction environment during concrete pouring and vibration. The installation direction of the strain sensors must be strictly consistent with the overall coordinate system of the support system to ensure that the three stress components measured subsequently accurately correspond to the coordinate directions in the finite element model. Specifically, the three measurement directions of the strain sensors are arranged along the vertical, the first horizontal direction of the support system, and the second horizontal direction of the support system, respectively. The first and second horizontal directions correspond to the longitudinal and transverse axes of the uprights in the planar layout diagram. All sensors should maintain a unified horizontal measurement direction during installation; that is, the first horizontal direction of all strain sensors should be along the same coordinate axis, and the second horizontal direction of all sensors should be along another coordinate axis. This ensures that the stress components throughout the field have a consistent directional reference during subsequent data processing, facilitating the construction of a spatially continuous stress field.

[0026] During the pouring of the base slab concrete, strain sensors are poured and formed synchronously with the concrete, creating a tight bonding interface between the sensors and the surrounding concrete, thus achieving coordinated deformation between the two. Once the base slab concrete has reached its final set and the specified curing age, the strain sensors are firmly anchored inside the base slab. When the concrete at the sensor location subsequently bears the load transmitted from the upper support system, the resulting minute strains can be effectively sensed and output through the strain sensor. Since a strain sensor is located directly beneath each upright, a regular array of strain sensors corresponding to the matrix arrangement of the uprights is formed throughout the entire base slab. Each measuring point in this array can simultaneously provide data on three independent strain components: vertical strain, first horizontal strain, and second horizontal strain, providing raw measurement information for the subsequent construction of the measured stress field and the three-channel stress deviation field.

[0027] In one technical solution, the strain sensor consists of a vertical reinforcing bar and three orthogonal strain gauges adhered to it, measuring strain in the vertical, first horizontal, and second horizontal directions respectively. The strain sensor is vertically fixed within the concrete base slab directly below each pole base, and all strain sensors have the same horizontal measurement direction. The vertical reinforcing bar can be ordinary building steel bars of the same or similar specifications as the base slab reinforcing bars, placed vertically along its length. Three strain gauges are fixed to the surface of the vertical reinforcing bar using an adhesive bonding process. The sensitive grid directions of the three strain gauges are arranged orthogonally to each other along the vertical, first horizontal, and second horizontal directions, thus forming a sensing unit capable of simultaneously measuring strain components in three orthogonal directions. After the strain gauges are bonded, the strain gauges and the welded lead wires are coated with multiple layers of epoxy resin or waterproof sealant for sealing. Each strain sensor's lead wire is tied along the reinforcing steel mesh of the base slab and converges at the centralized acquisition box at the edge of the base slab. The lead wire is entirely embedded inside the base slab, without protruding from the top surface, thus avoiding any interference with subsequent construction processes such as the erection of support poles and the laying of formwork. The strain sensors are fixed to the reinforcing steel mesh within the base slab using a binding method, ensuring their tops are directly below the designated support pole bases. All sensors must be uniformly calibrated during installation, ensuring that their first horizontal strain gauges point to the same coordinate axis of the support system, and their second horizontal strain gauges point to a different coordinate axis, thereby guaranteeing the consistency of horizontal measurement directions throughout the site. Constructing a triaxial strain sensor using strain gauges bonded to a reinforcing steel carrier significantly reduces costs compared to embedding prefabricated multiaxial strain gauges or stress gauges, making it suitable for applications requiring a large number of strain sensors under full-span scaffolding.

[0028] In one technical solution, identical curing test blocks are fabricated simultaneously with the pouring of the base slab concrete, and strain compensation sensors are embedded within these test blocks. Once the base slab concrete reaches its design strength, all strain sensors and the strain compensation sensors are simultaneously zeroed. In subsequent strain data processing, the measured strain values ​​of each strain sensor are subtracted from the free strain measured by the strain compensation sensors at the same time to deduct the influence of concrete shrinkage and temperature effects on stress calculation. Simultaneously with the pouring of the base slab concrete, several identical curing test blocks are fabricated using the same batch of concrete, and identical strain compensation sensors as those in the base slab are embedded within these test blocks. The identical curing test blocks and the base slab concrete are cured in the same environment, experiencing the same temperature and humidity conditions. After the base slab concrete reaches its design strength, before the erection of the support system, all strain sensors in the base slab and the strain compensation sensors in the identical curing test blocks are simultaneously zeroed to eliminate strain accumulation caused by concrete hardening shrinkage and temperature changes beforehand. During the subsequent pouring and curing of the top slab concrete, the measured strain values ​​from each strain sensor on the bottom slab and the free strain values ​​measured by the strain compensation sensor were collected simultaneously. During data processing, the strain components measured by the strain sensor under each pole in each direction were subtracted from the free strain measured by the strain compensation sensor in the corresponding direction at the same time. The resulting difference was considered to be the stress strain caused solely by the upper load transmitted by the support system. This strain subtraction process effectively reduced the systematic errors introduced by concrete shrinkage and temperature effects in stress back-calculation, making the final measured stress field of the bottom slab more accurately reflect the mechanical response under the pole load.

[0029] S2. Establish a finite element model that includes the base plate, support system and top load, simulate the stress distribution in the base plate under load, extract the normal stress in three orthogonal directions at the corresponding positions of the strain sensors, and form the theoretical stress field of the base plate. The finite element model can be established using general finite element analysis software, such as ABAQUS, ANSYS or MIDAS GEN, etc. Those skilled in the art can choose flexibly according to engineering habits and software conditions.

[0030] In the geometric modeling stage, a comprehensive model including the base slab concrete, support system uprights, and top construction loads needs to be established. The base slab concrete portion can be modeled using 3D solid elements according to the actual design thickness and planar dimensions. The mesh size of the elements needs to be appropriately refined near the sensor embedding locations to ensure the accuracy of stress extraction. For the support system portion, since the core focus of this invention is on the stress response within the base slab, the modeling of the full-span frame can adopt a reasonable simplification strategy. Specifically, uprights and crossbars can be simulated using spatial beam elements and line elements. The node connection form depends on the actual fastener, cup-lock, or disc-lock connection nodes used, and stiffness reduction can be performed as semi-rigid nodes, or envelopment analysis can be directly performed using hinged or rigid connections. Structural members such as scissor braces can be appropriately simplified or ignored if they have a small impact on the stress of the base slab, in order to reduce the model size and computation time. The loads of the top formwork, construction personnel, equipment, and poured concrete can be equivalent to uniformly distributed surface loads or applied through concentrated forces at the top of the uprights.

[0031] Regarding the definition of material properties, the concrete of the foundation slab is assigned the corresponding elastic modulus and Poisson's ratio according to the design strength grade, and the steel pipes in the support system are assigned the elastic modulus according to Q235 or Q345 steel. Since this scheme does not involve plastic deformation, it mainly focuses on the instantaneous elastic stress response under load, and the analysis type can adopt linear elastic static analysis, without the need to introduce material nonlinearity. As for boundary conditions, the bottom of the foundation slab is set with elastic foundation constraints or treated as a fixed boundary according to the actual foundation conditions, while the perimeter of the foundation slab is set with corresponding displacement constraints according to its connection relationship with the surrounding structure.

[0032] After the finite element model is established and solved, the stress extraction process begins. In the post-processing module of the finite element software, the theoretical normal stress components in three orthogonal directions are extracted at each strain sensor node location according to the actual embedded coordinates of the strain sensors within the base plate. These components are the theoretical vertical normal stress, the theoretical first horizontal normal stress, and the theoretical second horizontal normal stress. The directions of these three normal stress components must strictly correspond to the measurement directions of the three strain gauges of the strain sensor in S1 to ensure the comparability of subsequent theoretical and measured data. The extracted triaxial normal stress data for all measuring points are arranged according to the planar coordinates of the measuring points, thus forming the theoretical stress field of the base plate under this load condition. This theoretical stress field reflects the theoretical distribution of stress at various points inside the base plate under the premise that the uprights are in an ideal healthy state and the load is uniformly transmitted, providing a benchmark reference for subsequent point-by-point deviation calculations between the measured stress field and the actual stress field.

[0033] S3. Erect the support system and complete the pouring of the top slab concrete. Collect strain component data from all strain sensors and convert the strain component data into the measured stress field of the bottom slab. After the bottom slab concrete reaches the design strength and the strain sensors are zeroed, erect the full-span support system layer by layer above the bottom slab according to the construction plan. The base of the uprights must be accurately positioned directly above each strain sensor to ensure that the load transfer path is consistent with the sensor layout. After the support system is erected, lay the top slab formwork, tie the top slab reinforcement, and then pour the top slab concrete. Throughout the entire process of pouring the top slab concrete and during the curing stage after pouring, continuously or periodically collect data from the strain sensor array using data acquisition equipment to obtain the measured strain component data in three orthogonal directions at the corresponding measuring point of each strain sensor, namely the measured vertical strain, the measured first horizontal strain, and the measured second horizontal strain. During the acquisition process, simultaneously record the free strain data of the strain compensation sensors in the test blocks cured under the same conditions. After subtracting the free strain in the corresponding direction from the measured strain components at each measuring point, the stress-strain components caused solely by the load of the support system are obtained. Based on the elastic modulus and Poisson's ratio of concrete, the generalized Hooke's law is used to convert them into normal stress components in three orthogonal directions: measured vertical normal stress, measured first horizontal normal stress, and measured second horizontal normal stress. The three-dimensional measured normal stress data of all measuring points are arranged according to the plane position of the measuring points, thus constituting the measured stress field of the base slab. This measured stress field accurately records the stress distribution inside the base slab under actual construction loads and the actual working state of the support system, including information on local stress anomalies caused by various defects in the uprights.

[0034] S4. Calculate the point-by-point deviation between the measured stress field and the theoretical stress field, generating a three-channel stress deviation field composed of stress deviation values ​​in three orthogonal directions. After obtaining the theoretical stress field established in S2 and the measured stress field obtained in S3, calculate the deviation for each corresponding measuring point in the two stress fields. Specifically, for each measuring point corresponding to a strain sensor, subtract the theoretical vertical normal stress from the measured vertical normal stress to obtain the vertical stress deviation value of the measuring point; subtract the theoretical first horizontal normal stress from the measured first horizontal normal stress to obtain the first horizontal stress deviation value; subtract the theoretical second horizontal normal stress from the measured second horizontal normal stress to obtain the second horizontal stress deviation value. After completing the three-dimensional deviation calculation for all measuring points, arrange them according to the planar coordinates of the measuring points to form a vertical stress deviation channel, a first horizontal stress deviation channel, and a second horizontal stress deviation channel. These three channels together constitute the three-channel stress deviation field. The three-channel stress deviation field, presented in a structured grid data format, preserves the spatial relative positional relationship between each measuring point. The data from the three channels can be analogous to the pixel values ​​of the three RGB color channels in a digital image, and can be directly used as spatial feature data for subsequent convolutional neural network models.

[0035] S5. For each pole, using a pre-trained convolutional neural network model, based on the three-channel stress deviation field data of each measuring point in a local area centered on the pole, the type of damage to the pole is diagnosed, and the diagnostic results are output. After obtaining the complete three-channel stress deviation field covering the entire support system's base plate in S4, each pole needs to be diagnosed for damage based on this. When a pole has damage such as bottom suspension, top support eccentricity, pole tilting, or horizontal connection failure, the way the pole transmits load to the base plate will deviate from the ideal state. This deviation will cause a characteristic redistribution of triaxial stress in the base plate area below and around the pole. In other words, different types and degrees of pole damage will form stress deviations with specific spatial distribution patterns within the base plate. Because there is a clear spatial correspondence between the measuring points within the base plate and the uprights, and the measuring points themselves are arranged in a regular matrix, a local window covering the upright to be diagnosed and its adjacent uprights is extracted. The three-channel stress deviation matrix within this local window contains sufficient information for determining the type of damage to the central upright. This local three-channel stress deviation matrix is ​​input into a pre-trained convolutional neural network model, which automatically extracts the implicit multi-channel spatial features to identify the type of damage to the central upright. This process establishes a direct mapping between the spatial pattern of abnormal stress in the base plate and the type of damage to the uprights, thus bypassing the complex intermediate steps of manually analyzing the force transmission path and stress distribution required in traditional methods.

[0036] like Figure 2 As shown, taking a 10×10 pole area in the support system of a basement roof slab of a certain project as an example, the support system adopts a full-span steel pipe scaffold with couplers. The longitudinal spacing of the poles is 0.9 meters, the transverse spacing is 0.9 meters, the erection height is 5.5 meters, and the base slab is a 300 mm thick cast-in-place reinforced concrete structure. During the base slab pouring stage, strain sensors were pre-embedded in the base slab concrete directly below each pole according to the pole layout plan. All strain sensors have been zeroed and are in normal working condition. After the support system is erected and the roof slab concrete is poured, the strain component data of all strain sensors are collected. After deducting the free strain and converting it into stress, the measured stress field is obtained. The deviation is calculated point by point with the theoretical stress field obtained in advance through the finite element model, and finally a three-channel stress deviation field covering the entire base slab area of ​​the support system is generated.

[0037] Centered on the current position (5th row, 5th column) within the support system plane, a local window covering the current upright and its adjacent uprights is extracted from the three-channel stress deviation field. The stress deviation values ​​of each measuring point within this local window in the vertical, first horizontal, and second horizontal directions are respectively taken as the values ​​of the upright in the three independent channels. The measuring point directly below the upright is the grid corresponding to the position in the three-channel stress deviation field. Centered on this measuring point, two rows of measuring points are extended outwards to extract a 5×5 local window. This window contains 25 measuring point positions, covering the base plate area extending outwards from the current upright by twice the upright spacing. The window size is selected based on the mechanical laws of stress redistribution in the base plate: when a single upright experiences displacement, lateral constraint failure, or other defects, the resulting three-dimensional stress deviation in the base plate is mainly concentrated within a radius of approximately twice the upright spacing, centered on the measuring point below the upright. Meanwhile, extending the distance to the two rings of poles can include the cross-effects of possible defects on the stress field of the central area, allowing the convolutional neural network to have the opportunity to encounter and distinguish the different spatial superposition patterns of defects on the central pole and those on adjacent poles during the learning process.

[0038] For each measuring point within the 5×5 local window, its vertical stress deviation value, first horizontal stress deviation value, and second horizontal stress deviation value are extracted. The vertical stress deviation values ​​of these 25 measuring points are then sequentially filled into a 5×5 matrix according to their actual row and column order in the plane, serving as the first channel data for this local window. The 25 first horizontal stress deviation values ​​are filled into a second 5×5 matrix using the same rule, serving as the second channel data. The 25 second horizontal stress deviation values ​​are filled into a third 5×5 matrix, serving as the third channel data. This results in a 5×5×3 three-channel stress deviation matrix. This matrix, in a structured grid format, completely preserves the spatial distribution and directional information of the three-dimensional stress deviations within the local area, and can be directly used as standardized input for subsequent convolutional neural network models.

[0039] The three-channel stress deviation matrix is ​​input into a pre-trained convolutional neural network model, which includes an input layer, a first convolutional layer, a second convolutional layer, a flattening layer, a fully connected layer, and an output layer.

[0040] The input layer is used to receive the three-channel stress deviation matrix. The received three-channel stress deviation matrix has a size of 5×5×3. The values ​​of the three channels at 25 spatial positions of the matrix correspond to the stress deviations in the vertical direction, the first horizontal direction, and the second horizontal direction.

[0041] The first convolutional layer is used to simultaneously extract local spatial features from the three channels of the three-channel stress bias matrix, outputting a first intermediate result. It is configured with 16 convolutional kernels, each 3×3×3 in size covering all three channels, with a stride of 1 and no zero-padding. Each convolutional kernel performs a 3D convolution operation on the input matrix: it slides across a 5×5 spatial region of the input matrix, covering a 3×3 local window containing 27 values ​​across the three channels each time. These values ​​are then multiplied element-wise with the 27 weights of the kernel and summed, followed by a bias term. Finally, the result is processed by the ReLU activation function to output an activation value. After a convolutional kernel traverses the entire input matrix, it produces a 3×3 activation map. The 16 convolutional kernels output 16 3×3 activation maps as the first intermediate result, with a size of 3×3×16. These activation maps represent 16 different local spatial features. For example, one convolutional kernel responds strongly to the pattern of "large positive vertical deviation at the center and large negative horizontal deviation", while another convolutional kernel is sensitive to the pattern of "stress gradient monotonically changing along a certain direction".

[0042] The second convolutional layer further combines features from the first intermediate result to output a second intermediate result. This second convolutional layer can be configured with 32 convolutional kernels, each 3×3×16 in size, with a stride of 1 and no zero-padding. This layer performs convolutions on the 16 channels of the first intermediate result, with each kernel spanning all 16 channels, performing a weighted summation of 16 values ​​(each position corresponding to one channel value) within a 3×3 spatial region. After a kernel traverses the 3×3×16 input, it outputs a 1×1 activation value because the spatial size is exactly 3×3. The 32 convolutional kernels output a total of 32 1×1 activation maps, called the second intermediate result, with a size of 1×1×32. This layer combines the primary features extracted by the first layer into more abstract composite features, such as detecting the suspended feature pattern of "central vertical stress release accompanied by symmetrical enhancement of peripheral horizontal stress."

[0043] The flattening layer unfolds all the feature maps in the second intermediate result sequentially and concatenates them into a one-dimensional vector, completing the transformation from a spatial grid structure to the input format of a fully connected layer. The 32 1×1 activation maps are sequentially concatenated into a one-dimensional vector of length 32. This step completes the transformation from a spatial grid structure to the input format of a fully connected layer.

[0044] A fully connected layer is used to integrate the one-dimensional vector to obtain the defect feature representation of the pole. The fully connected layer contains several layers of neurons, which perform layer-by-layer weighted mapping and nonlinear transformation on the flattened one-dimensional vector, gradually integrating and compressing the information scattered in various feature maps, and finally obtaining a highly condensed one-dimensional vector, which is the defect feature representation of the current pole. Each dimension of the vector encodes abstract information related to a certain aspect of the pole defect in the input three-channel stress deviation matrix.

[0045] The output layer is used to output the probability that the pole belongs to each disease type based on the disease characteristics, and to calculate the probability value of the current pole belonging to each predefined disease type through a normalized exponential function.

[0046] The spatial distribution characteristics of the three-dimensional stress deviation in the three-channel stress deviation matrix are extracted by the convolutional neural network model, and the defect type of the pole is output accordingly. The defect type includes at least one of the following: pole bottom suspension, top support eccentricity, pole tilting, and horizontal connection failure.

[0047] The convolutional neural network model extracts and integrates features layer by layer from the input 5×5×3 matrix, resulting in a probability distribution at the output layer for five categories of the pole: bottom suspension, top eccentricity, pole tilting, horizontal connection failure, and normal state. The category with the highest probability value is taken as the diagnostic conclusion for the pole, indicating that it has a bottom suspension defect. If the probability values ​​of each category in the model's output probability distribution are relatively even, without a clearly prominent category, the pole can be considered to be in a normal state or the defect characteristics are not yet obvious.

[0048] In the convolutional neural network model structure of the technical solution, no pooling layer is set. This design is based on specific considerations of the size characteristics of the input data and the task requirements. In conventional convolutional neural networks, pooling layers are usually used to reduce the spatial size of feature maps, reduce the number of parameters, and introduce a certain degree of translation invariance. However, its essence is to expand the receptive field of subsequent layers at the cost of spatial resolution through downsampling operations. However, in this solution, the three-channel stress deviation matrix input to the convolutional neural network is itself relatively small. After the convolution operations of the first and second convolutional layers, the size of the feature map has been further reduced. If a pooling layer is introduced in this process, it will further compress the spatial dimension of the feature map, which may lead to a serious loss of the already limited spatial information. This is not conducive to preserving the fine relative positional relationship and gradient change details between the stress deviation values ​​of each measuring point. These details are precisely the key clues for distinguishing different types of defects. For example, the spatial contrast pattern of reduced vertical stress at the central measuring point and increased stress at the surrounding measuring points caused by the suspension of the bottom of the pole needs to retain the spatial resolution at the measuring point level to be accurately identified. Therefore, this scheme, by discarding the pooling layer and retaining the spatial accuracy of the feature map, enables the entire network to extract and transmit the spatial distribution characteristics of local stress anomalies caused by pole defects more completely, thereby ensuring the accuracy of classification and regression tasks.

[0049] After completing the diagnosis of the current pole, slide the center of the local window to the measurement point position corresponding to the next pole, repeat the above process of extracting the matrix, inputting the model, and outputting the diagnosis, traversing the entire support system pole by pole, and finally obtaining the independent disease diagnosis conclusion for each pole.

[0050] S6. Summarize the diagnostic results of all uprights, mark the locations of abnormal uprights, and generate reinforcement plans. After completing the diagnosis of each upright in the support system, summarize and organize the diagnostic results for each upright. The diagnostic results include the planar coordinates of the upright, the type of defect, and the probability value of each defect type. Based on the summarized results, mark all uprights identified as having defects on the support system layout plan, using different colors or symbols to distinguish the defect types. For example, red marks the position where the bottom of the upright is suspended, yellow marks the position where the top is eccentric, and blue marks the position where the upright is tilted, thus forming a visual defect distribution map. For the marked abnormal uprights, retrieve the corresponding treatment plan from the preset reinforcement measure library according to their defect type and the stress characteristics of the area. For example, for defects such as the bottom of the upright being suspended, it is recommended to fill the suspended area with steel plates or to re-install the uprights around the perimeter; for defects such as eccentric top supports, it is recommended to adjust the position of the top supports and retighten them; for defects such as upright tilting or horizontal connection failure, it is recommended to add horizontal tie rods or scissor braces for reinforcement. The reinforcement plan should be considered in conjunction with the distribution of the defects. Figure 1 The data is then output to allow on-site construction personnel to follow the diagrams and address each point systematically, ensuring that the support system remains in a safe and reliable working state under subsequent construction loads.

[0051] In another technical solution, the output layer further includes a regression output branch, which outputs multiple disease severity parameters. The convolutional neural network model selects effective parameters corresponding to the disease type from the multiple disease severity parameters based on the classification results. When the value of the effective parameter exceeds a preset severity threshold, the disease on the pole is immediately treated on-site. When constructing the regression output branch, a mature multi-task learning network structure in existing technologies can be referenced, namely, an architecture mode of "hard parameter sharing" plus "multiple output heads". Specifically, the regression output branch and the classification output branch share the same disease feature representation vector output by the fully connected layer, but an independent fully connected sub-network parallel to the classification output branch is set within the output layer, forming two output heads with different functions. The classification output head is the normalized exponential function layer, responsible for outputting the probability of each category; the regression output head consists of one or more fully connected layers, with no activation function or a linear activation function at its end, directly outputting continuous values, each value corresponding to a disease severity parameter.

[0052] In another technical solution, the training process of the convolutional neural network model includes: Define a joint loss function, which is a weighted sum of a classification loss term and a regression loss term. The classification loss term measures the difference between the predicted value of the disease type and the true label, and the regression loss term measures the difference between the predicted value of the disease severity parameter and the true label. Train the convolutional neural network model to be trained using a pre-constructed training sample set. Iterate the network weights by minimizing the joint loss function and using the backpropagation algorithm until the model converges.

[0053] Convolutional neural network models simultaneously perform two tasks: disease type classification and severity regression. During training, a joint loss function needs to be defined to guide both learning objectives. This joint loss function consists of a classification loss term and a regression loss term, weighted by preset weighting coefficients. The classification loss term measures the difference between the probability distributions of each category output by the model and the true disease type labels, and can be calculated using the cross-entropy loss function. Specifically, when the true label of a training sample is "bottom of the pole is suspended," the classification loss term quantifies the deviation between the probability value of the "bottom suspended" category in the model's output probability distribution and the ideal value of 1; the larger the deviation, the higher the loss value. The regression loss term measures the difference between multiple severity parameters output by the model's regression output branch and the true severity labels, and can be calculated using the mean squared error loss function or the mean absolute error loss function. In actual calculation of the regression loss term, for each training sample, only the prediction error of the severity parameter corresponding to the true disease type of that sample is calculated; other severity parameters unrelated to the current disease type are not included in the regression loss calculation. For example, if the true label of a training sample is top support eccentricity with an eccentricity of 5 mm, then the regression loss term only calculates the difference between the top support eccentricity parameter output by the model and 5 mm, while parameters such as the degree of suspension at the bottom of the pole and the tilt angle of the pole do not contribute to the regression loss in this sample. The joint loss value for this training sample is obtained by multiplying the classification loss term and the regression loss term by their respective weight coefficients and then summing them.

[0054] The training process utilizes a pre-constructed large-scale training sample set, the construction method of which is further explained below. In each training iteration, a batch of training samples is sequentially input into the convolutional neural network model to be trained. The model calculates layer by layer according to the forward propagation process and finally outputs classification prediction results and regression prediction results. Based on the predicted values ​​output by the model and the true values ​​of the sample labels, the average joint loss value of the current batch is calculated according to the joint loss function mentioned above. Subsequently, the gradient of the joint loss value with respect to the weights of each layer in the network is calculated layer by layer from the output layer using the backpropagation algorithm, and the weight parameters of each layer are updated according to the gradient descent optimization algorithm. The above forward propagation, loss calculation, backpropagation, and weight update steps are repeated, traversing the training sample set multiple times, until the joint loss value drops below a preset threshold or tends to stabilize and no longer decreases significantly. At this point, the model is considered to have converged. The converged model weight parameters are then fixed and saved for subsequent online diagnosis of pole defects in actual engineering projects.

[0055] First, using the finite element model established in S2, which includes the base plate, support system, and top load, as the basic template, the parametric modeling function of the finite element software is used to characterize various defects of the uprights as controllable model parameters. Specifically, the defect of the upright bottom being suspended can be simulated by setting a gap distance parameter between the bottom node of the upright and the base plate. This gap distance parameter is randomly selected within a preset range when generating different samples to cover different degrees of suspension, from slight to severe. The defect of top support eccentricity can be simulated by applying an eccentricity parameter to the top of the upright. The magnitude and direction of the eccentricity can be used as variables. The defect of upright tilting can be simulated by applying an initial tilt angle parameter to the upright. Both the tilt angle and the tilt direction can be adjusted. The defect of horizontal connection failure can be simulated by weakening or releasing the rotational stiffness parameter of the connection node between the upright and the crossbar. The stiffness reduction coefficient varies within a preset range. By adjusting different combinations of the above parameters, a large number of finite element model samples with different defect types and severity can be derived from the same basic model.

[0056] For each parameterized finite element model, a top construction load consistent with the actual situation is applied for static solution. After the solution is completed, the triaxial stress deviation field of the model is obtained using the same method as in S4. It is important to note that during the sample generation stage, the theoretical stress field is uniformly based on the calculation results of an ideal healthy model without any defects, while the measured stress field is replaced by the calculation results of each defective model. The difference between the two is obtained by point-by-point subtraction at the measurement points to obtain the triaxial stress deviation field under the defective state. Subsequently, using the same local window extraction method as in S5, a local area covering the vertical pole and its adjacent vertical poles is extracted from the triaxial stress deviation field, centered on each pole, to construct a three-channel stress deviation matrix with dimensions matching the network input layer. Since the defective state of each vertical pole in the finite element model is manually set, the defect type and defect severity parameters corresponding to the three-channel stress deviation matrix are known quantities, and these known quantities are directly labeled as the sample. Each training sample consists of a three-channel stress deviation matrix and a set of label data, which includes discrete disease type codes and continuous severity parameter values. By automating the above parameter modification, model solving, stress extraction, and label annotation process in a loop, tens of thousands of training samples can be generated efficiently, covering various disease types, severity levels, and complex working conditions with multiple coupled diseases. This provides a sufficient, diverse, and accurately labeled sample data foundation for the full training of the convolutional neural network model.

[0057] A monitoring system for tall formwork support systems, applied to the aforementioned monitoring method for tall formwork support systems, comprises three main components: a strain sensor array, acquisition equipment, and data processing equipment.

[0058] The strain sensor array is embedded in the concrete base slab directly beneath each pole. This array consists of several triaxial strain sensors arranged in a planar matrix according to the poles, with each sensor corresponding to one pole. The sensor body uses vertical reinforcing steel as a carrier; the steel can be HRB400 grade construction steel with a diameter of 16 mm or 20 mm, and its length is adapted to the thickness of the base slab, typically the net height range of the base slab thickness minus the thickness of the upper and lower protective layers. A resistance strain gauge is attached to the surface of the reinforcing steel along the vertical and two orthogonal horizontal directions. The strain gauges can be of the BE120-3AA type from AVIC Electromechanical Measurement or the KFGS series general-purpose foil strain gauges from Kyowa Corporation of Japan, with a gauge length of 3 mm to 5 mm and a resistance value of 120 ohms or 350 ohms. After attachment, the strain gauges are protected with multiple layers of epoxy resin or polyurethane sealant. The leads are four-core shielded cables, tied along the reinforcing steel mesh of the base slab and converged to a terminal box at the edge of the base slab. All sensor leads are embedded inside the base plate and not exposed on the top surface of the base plate, so as not to affect subsequent pole erection and formwork laying operations.

[0059] The data acquisition device is electrically connected to the strain sensor array and is used to collect strain data from each strain sensor during and after the pouring of the roof slab concrete. The acquisition device can be a static strain testing system such as the DH3816N model from Jiangsu Donghua Testing or the uT7110Y model from Wuhan Youtai. This system supports multi-channel synchronous acquisition, with each acquisition module typically supporting 16 or 32 channels. The modules can be expanded and networked according to the total number of sensors. The acquisition device is connected to the sensor leads via a terminal box, using a 1 / 4 bridge, half-bridge, or full-bridge connection. In this solution, each strain gauge operates independently; a 1 / 4 bridge connection with temperature compensation elements is recommended. The acquisition device can communicate with the data processing equipment via a USB or Ethernet interface, supporting real-time continuous sampling or timed sampling modes. The sampling frequency can be set from once every 1 minute to 10 minutes according to the construction progress, and can be increased to once per second during critical stages of concrete pouring. The acquisition device should be placed in a rainproof and dustproof protective box at the construction site and equipped with an uninterruptible power supply to ensure continuous data recording throughout the pouring process.

[0060] A data processing device, communicatively connected to the acquisition device, can be an industrial-grade laptop, industrial PC, or tower server, such as an Advantech IPC-610 series industrial PC or a Dell PowerEdge series server, and is deployed in a temporary on-site monitoring room. The data processing device comprises three functional modules: a stress field generation module, a diagnostic module, and an early warning module, each implemented in software. The data processing device includes: The stress field generation module converts the acquired strain data into a measured stress field and calculates the deviation from the pre-stored theoretical stress field point by point to generate a three-channel stress deviation field. This module first reads the triaxial strain component data from each measuring point from the acquisition device, and simultaneously reads the free strain data measured by the strain compensation sensor in the same cured specimen. The two are subtracted according to their corresponding directions to obtain the stress-strain components. Then, based on the measured elastic modulus and Poisson's ratio of the base slab concrete, the triaxial stress-strain components are converted into triaxial normal stress components according to the generalized Hooke's law. The stress components of all measuring points are arranged in planar positions to form the measured stress field. This module internally stores the theoretical stress field data file obtained from the finite element model calculation in step S2. The measured stress field is subtracted from the theoretical stress field at each measuring point to obtain the vertical stress deviation field, the first horizontal stress deviation field, and the second horizontal stress deviation field. The three deviation fields are combined and output as the three-channel stress deviation field. This module can be written and implemented in platforms such as MATLAB, Python's NumPy and SciPy libraries, or LabVIEW, and supports batch data import, matrix operations, and visualization.

[0061] The diagnostic module, with its built-in pre-trained convolutional neural network model, outputs the type of damage to each pole based on the three-channel stress deviation field data of each measuring point within a local area centered on that pole. During operation, the module extracts a 5×5 local window centered on the current pole from the three-channel stress deviation field provided by the stress field generation module, constructing the corresponding three-channel stress deviation matrix for each pole in sequence. It then calls the pre-trained convolutional neural network model file for inference calculations. The model file can be stored in the format of deep learning frameworks such as TensorFlow, PyTorch, or ONNX. The diagnostic module analyzes the probability distribution of the damage type output by the model and the severity parameters output by the regression branches to determine the most likely damage category and severity for the current pole, storing the diagnostic conclusions pole by pole in the results database. The diagnostic module can be developed using Python combined with a deep learning framework, and the inference calculations can be accelerated with GPUs to improve processing efficiency, such as using NVIDIA Tesla T4 or RTX series graphics cards.

[0062] The early warning module summarizes the diagnostic results of all poles, marks the locations of abnormal poles, and generates reinforcement plans. After obtaining the diagnostic results from the diagnostic module, this module first counts the number and spatial distribution of poles with various defects. Abnormal poles and their defect types are then marked with different colors and symbols on the support system layout plan, creating a visual defect distribution map. Poles with severity parameters exceeding a preset threshold are highlighted with flashing indicators or bold borders. Subsequently, based on the defect type and severity, the module matches corresponding reinforcement plan text from a preset reinforcement measures database. For example, bottom suspension corresponds to "filling with wedge-shaped steel plates and injecting high-strength grout," and top support eccentricity corresponds to "loosening the top support, realigning, and then tightening." The plan is then matched one-to-one with the pole location, generating a detailed reinforcement list containing the pole number, location, defect type, severity, and recommended treatment measures. This module can be developed using languages ​​such as C# and Java, supports exporting results as PDF reports or Excel spreadsheets, and can push alarm information to the handheld terminals of on-site management personnel via LAN or 4G / 5G network for immediate response.

[0063] A computer device includes a memory and a processor. The memory stores computer instructions, and the processor executes the computer instructions to perform the above-described high-performance template support system monitoring method. The computer device can be any terminal device, including mobile phones, laptops, desktop computers, tablets, PDAs (Personal Digital Assistants), POS (Point of Sales), in-vehicle computers, etc.

[0064] Computer instructions are stored in a readable storage medium, such as a computer floppy disk, USB flash drive, portable hard drive, read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0065] It should be noted that although the steps are described in a specific order above, this does not mean that they must be performed in that order. In fact, some of these steps can be executed concurrently, or even in a different order, as long as the required functionality is achieved. The number of devices and processing scale described herein are for simplification of the invention; applications, modifications, and variations of this invention will be readily apparent to those skilled in the art.

[0066] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A monitoring method for tall formwork support systems, characterized in that, Includes the following steps: S1. Based on the planar arrangement of the support system poles, a strain sensor is pre-embedded in the base plate concrete directly below each pole. The strain sensor is used to measure the strain components in three orthogonal directions within the concrete. S2. Establish a finite element model including the base plate, support system and top load, simulate the stress distribution in the base plate under load, and extract the normal stress in three orthogonal directions at the corresponding positions of the strain sensor to form the theoretical stress field of the base plate. S3. Erect the support system and complete the concrete pouring of the top slab. Collect the strain component data of all strain sensors and convert the strain component data into the measured stress field of the bottom slab. S4. Calculate the point-by-point deviation between the measured stress field and the theoretical stress field, and generate a three-channel stress deviation field composed of stress deviation values ​​in three orthogonal directions. S5. For each pole, using a pre-trained convolutional neural network model, based on the three-channel stress deviation field data of each measuring point in the local area centered on the pole, the type of damage to the pole is diagnosed, and the diagnosis result is output. S6. Summarize the diagnostic results of all poles, mark the locations of abnormal poles, and generate reinforcement plans.

2. The monitoring method for tall formwork support systems as described in claim 1, characterized in that, The strain sensor consists of a vertical steel bar and three orthogonal strain gauges attached thereon, which measure the strain in the vertical, first horizontal, and second horizontal directions, respectively. The strain sensor is vertically fixed in the concrete base plate directly below the base of each pole, and the horizontal measurement direction of all strain sensors is the same.

3. The monitoring method for tall formwork support systems as described in claim 1, characterized in that, Simultaneously prepare test blocks for curing under the same conditions when pouring the base slab concrete, and embed strain compensation sensors in the test blocks for curing under the same conditions. Once the base slab concrete reaches its design strength, all strain sensors and the strain compensation sensor will be simultaneously reset to zero. In subsequent strain data processing, the measured strain values ​​of each strain sensor are subtracted from the free strain measured by the strain compensation sensor at the same time to deduct the influence of concrete shrinkage and temperature effects on stress calculation.

4. The monitoring method for tall formwork support systems as described in claim 1, characterized in that, Step S5 includes: Centered on the current pole, a local window covering the pole and its adjacent poles is extracted from the three-channel stress deviation field. The stress deviation values ​​of each measuring point in the local window in the vertical, first horizontal and second horizontal directions are respectively used as the values ​​of the pole in the three independent channels. The measuring points are arranged according to their planar positions to form the three-channel stress deviation matrix of the local window. The three-channel stress deviation matrix is ​​input into a pre-trained convolutional neural network model; The spatial distribution characteristics of the three-dimensional stress deviation in the three-channel stress deviation matrix are extracted by the convolutional neural network model, and the defect type of the pole is output accordingly. The defect type includes at least one of the following: pole bottom suspension, top support eccentricity, pole tilting, and horizontal connection failure.

5. The monitoring method for tall formwork support systems as described in claim 4, characterized in that, The convolutional neural network model includes: The input layer is used to receive the three-channel stress deviation matrix; The first convolutional layer is used to simultaneously extract local spatial features from the three channels of the three-channel stress deviation matrix and output the first intermediate result. The second convolutional layer is used to further combine features from the first intermediate result and output the second intermediate result. A flattening layer is used to stretch the second intermediate result into a one-dimensional vector; A fully connected layer is used to integrate the one-dimensional vector to obtain a representation of the disease characteristics of the pole. The output layer is used to output the probability that the pole belongs to each type of disease based on the disease characteristics.

6. The monitoring method for tall formwork support systems as described in claim 5, characterized in that, The output layer also includes a regression output branch, which is used to output multiple disease severity parameters. The convolutional neural network model selects effective parameters corresponding to the disease type from the multiple disease severity parameters according to the classification results. When the value of the effective parameter exceeds the preset severity threshold, the disease of the pole is immediately treated on-site.

7. The monitoring method for tall formwork support systems as described in claim 6, characterized in that, The training process of the convolutional neural network model includes: Define a joint loss function, which is composed of a weighted sum of a classification loss term and a regression loss term. The classification loss term is used to measure the difference between the predicted value of the disease type and the true label, and the regression loss term is used to measure the difference between the predicted value of the disease severity parameter and the true label. The convolutional neural network model to be trained is trained using a pre-built training sample set. The network weights are iteratively updated by minimizing the joint loss function and using the backpropagation algorithm until the model converges.

8. The monitoring method for tall formwork support systems as described in claim 7, characterized in that, The construction of the training sample set includes: Training samples are generated by parametric finite element model. Different types of pole defects are set in the finite element model. After applying construction load, the triaxial stress deviation field of the area around the pole is extracted. A three-channel stress deviation matrix is ​​constructed and the corresponding defect type and defect severity parameters are labeled as tags. Each training sample contains a three-channel stress bias matrix and its corresponding label.

9. A monitoring system for a tall formwork support system, applied to the monitoring method for a tall formwork support system as described in claims 1-8, comprising: The strain sensor array is embedded in the concrete base plate directly below each pole; The data acquisition device is electrically connected to the strain sensor array and is used to acquire strain data from each strain sensor during and after the pouring of the top slab concrete. A data processing device, communicatively connected to the acquisition device, the data processing device comprising: The stress field generation module is used to convert the collected strain data into a measured stress field and calculate the deviation from the pre-stored theoretical stress field point by point to generate a three-channel stress deviation field. The diagnostic module has a built-in pre-trained convolutional neural network model, which is used to output the type of damage to the pole based on the three-channel stress deviation field data of each measuring point in a local area centered on each pole. The early warning module is used to summarize the diagnostic results of all poles, mark the locations of abnormal poles, and generate reinforcement plans.

10. A computer device, characterized in that, The system includes a memory and a processor, wherein the memory stores computer instructions, and the processor executes the computer instructions to perform the monitoring method for the tall template support system as described in any one of claims 1 to 7.

Images