CN-121998253-A - Dynamic monitoring method for building carrier plate construction process

CN121998253ACN 121998253 ACN121998253 ACN 121998253ACN-121998253-A

Abstract

The application relates to the technical field of artificial intelligence, in particular to a dynamic monitoring method for a construction process of a floor support plate, which comprises the steps of collecting stress data of the whole construction period of the floor support plate, constructing a training data set with a construction stage type label, executing self-adaptive multi-scale morphological filtering processing on the stress data in the training data set, respectively processing the stress data by adopting two compound operations of executing a closing operation after a morphological opening operation and executing an opening operation after the closing operation based on a pre-defined discrete structural element scale set, constructing a construction monitoring model and utilizing the training data set to complete model training so as to dynamically monitor the construction process, wherein the construction monitoring model takes a deep convolutional neural network with self-adaptive depth convolution and cross-layer characteristic bridging as a trunk characteristic extractor, and determines the classification probability of the construction stage through a weighted loss function based on an attention classification layer guided by external characteristics and mutation perception.

Inventors

Xia Handuo
ZHANG KUNLIANG
ZHAO SHUAI
OuYang Sinan
LIU ZHENXING
QU CHENCHEN
ZHAO YAN
ZUO BIN
JIA FU
ZHOU FENG
ZHANG XIANG
Ruan Yingxin

Assignees

二十二冶集团雄安发展有限公司
中国二十二冶集团有限公司

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. The dynamic monitoring method for the construction process of the floor support plate is characterized by comprising the following steps of: Collecting stress data of the whole construction period of the floor support plate, and constructing a training data set with a construction stage type label; The adaptive multi-scale morphological filtering processing is carried out on the stress data in the training data set, and comprises the steps of predefining a discrete structural element scale set, adopting two compound operations of executing a closing operation after morphological opening operation and executing an opening operation after closing operation to respectively process the stress data, determining a dominant scale by analyzing residual energy distribution of the stress data and two compound operation results under each scale; And constructing a construction monitoring model and completing model training by utilizing the training data set to dynamically monitor the construction process, wherein the construction monitoring model takes a depth convolution neural network with self-adaptive depth convolution and cross-layer feature bridging as a trunk feature extractor, and determines the classification probability of a construction stage through a weighted loss function of an attention classification layer and abrupt change perception guided by external features.
2. The method for dynamically monitoring the construction process of the floor support plate according to claim 1, wherein the collected stress data comprises a stress time sequence signal, a corresponding time stamp and a construction activity mark; the construction of the training data set with construction stage category labels comprises the following steps: Cutting continuous stress data into a plurality of data segments with fixed length or corresponding to construction activity duration; According to the main construction activity in each data segment acquisition time period, corresponding construction stage type labels are given to each data segment, wherein the construction stage type labels comprise a construction preparation stage, a concrete pouring stage, a curing hardening stage and a loading test stage; and constructing the training data set by the data segment with the construction stage category label.
3. The method for dynamically monitoring the construction process of a floor support plate according to claim 1, wherein the stress data is processed by two complex operations of performing a closing operation after a morphological opening operation and performing an opening operation after the closing operation, respectively, comprising: constructing a self-adaptive multi-scale morphological filtering frame, and respectively performing opening and closing operation on the acquired original stress data by utilizing structural elements; and carrying out weighted fusion on the two operation results according to the self-adaptive weights meeting the complementary relation to obtain the filtering stress data vector for filtering the positive and negative impulse noise.
4. The method for dynamically monitoring the construction process of a floor support plate according to claim 1, wherein the open/close operation result is determined, the method comprising: Adopting flat linear structural elements under dominant scale to respectively execute opening and closing operation on original stress data to obtain an opening and closing operation result and a closing operation result; The self-adaptive weights comprise self-adaptive weights corresponding to an opening and closing operation result and self-adaptive weights corresponding to a closing operation result, and are determined based on residual vectors under a dominant scale, wherein the residual vectors comprise first residual vectors of original stress data and the opening and closing operation result under the dominant scale, and second residual vectors of the original stress data and the closing operation result under the dominant scale.
5. The method for dynamically monitoring the construction process of the floor support plate according to claim 1, wherein the construction monitoring model further comprises a dual-path feature extraction module for extracting a global contour feature vector and a local significant fluctuation feature vector of the filtered stress data, and fusing the global contour feature vector and the local significant fluctuation feature vector to obtain the input of the adaptive deep convolution.
6. The method for dynamically monitoring a floor support plate construction process according to claim 5, wherein the dual-path feature extraction module comprises a first path and a second path; The first path is used for carrying out self-adaptive segmentation on the filtered stress data, calculating weighted features based on linear fitting parameters of all segments and splicing to obtain the global contour feature vector; The second path is used for extracting overlapped subsequence fragments from the filtered stress data, coding the fragments through a pre-trained leachable sparse coding dictionary, and counting the mean value and standard deviation of coding coefficients to obtain the local significant fluctuation feature vector; The dual-path feature extraction module is used for splicing the global contour feature vector and the local significant fluctuation feature vector, and then obtaining an enhanced feature vector through linear transformation layer fusion, and the enhanced feature vector is used as the input of the self-adaptive depth convolution.
7. The method for dynamically monitoring the construction process of the floor support plate according to claim 1, wherein convolution blocks are configured in the deep convolution neural network, each convolution block comprises self-adaptive deep convolution, batch normalization, activation functions and cross-layer characteristic bridging, the self-adaptive deep convolution introduces a channel attention mechanism, and channel attention weights are calculated through global average pooling and full-connection layers so as to modulate the contribution degree of the convolution of each channel; and the cross-layer feature bridging adjusts the dimension of the shallow feature map through one-dimensional 1 multiplied by 1 convolution with the step length of 2, and is added with the current layer features to realize fusion.
8. The method for dynamically monitoring the construction process of the floor carrier plate according to claim 1 is characterized in that the construction monitoring model processes the output feature map of the deep convolutional neural network through an attention classification layer guided by external features, fuses statistical features of filtered data, and outputs classification probability through a full connection layer and a Softmax function; The external feature vector is determined based on a statistical feature vector obtained by calculating a filtering stress data vector; The key transformation weight matrix and the value transformation weight matrix are respectively used for mapping the characteristic vector of each time position of the stress data into a key vector and a value vector, and the attention calculation process is also related with the dimension of the query vector and the key vector, and the weight matrix and the bias vector of the classification output layer.
9. The method of claim 1, wherein training the construction monitoring model further comprises: The model parameters are iteratively updated by adopting a self-adaptive moment estimation optimization algorithm, and model training is completed by utilizing a total loss function, wherein the total loss function is formed by weighting and summing a weighted cross entropy loss of abrupt perception and a attention regularization loss through a regularization coefficient; The weighted cross entropy loss of abrupt change perception is determined based on preset abrupt change weight coefficients, average gradient amplitude of filtering stress data vectors, construction stage class labels and elements of prediction classification probability vectors; the attention regularization loss is determined based on attention weights corresponding to time locations of the stress data.
10. The method for dynamically monitoring the construction process of a floor support plate according to claim 1, wherein the construction process monitoring based on the construction monitoring model comprises the following steps: Dividing the real-time filtered data by adopting a sliding window mode, inputting the data into a construction monitoring model to obtain various probabilities, and outputting the class with the highest probability as a current construction stage identification result; When an abnormal stress pattern or phase deviation is identified, an early warning signal is triggered.

Description

Dynamic monitoring method for building carrier plate construction process Technical Field The application relates to the technical field of artificial intelligence, in particular to a dynamic monitoring method for a building carrier plate construction process. Background In the construction process of the floor support plate, construction activities such as vibration and loading can lead the floor support plate to generate complex stress response, and key information reflecting construction quality and structural safety is contained in the stress signals. Therefore, the stress signals in the construction process are monitored in real time and accurately analyzed, effective identification of the construction stage is achieved, and the method is an important means for guaranteeing the construction quality of the floor carrier plate and avoiding the structural safety risk. At present, the related construction monitoring field gradually introduces the technologies such as signal processing, feature engineering, machine learning and the like to try to improve the intelligent level of monitoring, but the existing monitoring method still has a plurality of technical defects to be solved urgently in practical application, and the high-precision monitoring requirement under the complex construction environment is difficult to meet. In the signal preprocessing link, the conventional frequency domain filtering and wavelet threshold denoising method is a mainstream signal denoising method at present, but when the method processes non-stationary stress signals generated by floor support plate construction, local abrupt change characteristics caused by construction activities such as vibration, loading and the like can be excessively smoothed, so that key engineering information is lost. In the feature extraction link, the conventional feature engineering method or the single-path automatic encoder is difficult to capture the macroscopic trend change and microscopic fluctuation detail of the stress signal at the same time, and the extracted features often have the problem of limited differentiation, so that the classification performance is affected. In the deep feature extraction and pattern recognition links, a standard convolutional neural network is often used for feature learning and stage recognition of stress signals, but the type of network adopts a convolutional kernel with fixed size, and an effective cross-layer connection mechanism is lacked, so that the multi-scale characteristics of construction stress signals cannot be adapted in a self-adaptive manner, and the depth feature extraction is insufficient. More importantly, the existing monitoring method has the common technical link fracture problem, namely, the signal processing, the characteristic engineering and the mode identification are regarded as mutually independent processes, and the prior knowledge of construction mechanics is not integrated into the whole algorithm design process. The generalization capability is insufficient when facing the complex interference of the site, and the false alarm rate is high. In conclusion, the existing floor support plate construction monitoring method has the problems of insufficient capture of key stress mutation characteristics, weak anti-interference capability, low stage identification precision, poor generalization capability and the like, and is difficult to meet the requirements of intelligent and self-adaptive monitoring in the actual construction process, so that the improvement of the floor support plate construction quality and the structural safety guarantee level is restricted. Disclosure of Invention In order to solve the technical problems, the application provides a dynamic monitoring method for a floor support plate construction process, which aims to realize accurate filtering and characteristic enhancement of a construction full-period stress signal, and finally achieves the purposes of automatic, high-precision and real-time identification and state tracking of key construction stages such as concrete pouring, vibrating, maintenance and loading by utilizing a neural network model embedded in engineering physical priori, thereby providing reliable data support and decision basis for construction quality control and safety management. The technical scheme provided by the application is as follows: a dynamic monitoring method for a building carrier plate construction process comprises the following steps: Collecting stress data of the whole construction period of the floor support plate, and constructing a training data set with a construction stage type label; the adaptive multi-scale morphological filtering processing is carried out on the stress data in the training data set, and comprises the steps of predefining a discrete structural element scale set, adopting two compound operations of executing a closing operation after morphological opening operation and executing an opening operation af