Wood structure damage detection and evaluation method based on physical information neural network

Abstract

The application discloses a wood structure damage detection and evaluation method based on a physical information neural network, which comprises the following steps: collecting ECDRM data of a wood structure surface by using a data acquisition device; inputting the preprocessed ECDRM data into a physical information neural network model to obtain the identification result of the damage type and degree of the wood structure; and displaying the identification result of the damage type and degree through a graphical interface. The method realizes comprehensive monitoring of wood structure damage through multiple links such as ECDRM data acquisition, model training and testing, result analysis and visualization, online monitoring and early warning, and provides strong support for practical application. Meanwhile, the application also has strong scalability and practicability, and can be applied to other types of structure damage detection fields.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of building detection, and particularly to a wood structure damage detection and evaluation method based on physical information neural network. BACKGROUND

[0002] Wood, as a natural and renewable building material, has high sustainability in the construction industry. Wood structure can maximize the green environmental performance of civil infrastructure, and has a lower carbon footprint compared with traditional building materials such as steel and concrete. However, wood structure is easily damaged by environmental factors such as humidity, insects and fungi, thereby affecting its structural integrity. Traditional detection methods such as visual inspection, probe and knocking method often have subjectivity and take a long time.

[0003] In recent years, the progress of electronic technology has promoted the development of new non-destructive testing (NDT) technologies, which can effectively detect the damage of wood structure. One of these technologies is drill resistance measurement (DRM), which involves drilling a hole into the wood and measuring the drill resistance. Drill resistance can be used to determine the strength and stiffness of the wood, thereby detecting the damage of the wood. Due to the drilling process, DRM is sometimes considered a semi-destructive technology.

[0004] Electronic controlled drill resistance measurement (ECDRM) is a more advanced form of DRM, which uses electronic control to adjust the speed and force of the drill needle, thereby achieving a more uniform feed speed, eliminating human error and variability in the testing process, and thus can provide more accurate and consistent results. However, analyzing and interpreting the data obtained from ECDRM tests can be challenging for non-experts. In addition, due to the complexity of the engineering and the presence of various variables in the testing environment, it becomes more difficult to accurately interpret the measured data in practical applications. SUMMARY

[0005] In view of the above defects of the prior art, the technical problems to be solved by the present application are the problems that the results of the existing electronic control drilling resistance measurement are difficult to apply and the problems that the existing wood structure damage detection technology is highly subjective and low in accuracy. The method of the present application combines a machine learning framework considering physical equation constraints (PINN) and a one-dimensional convolutional neural network (1D-CNN) for processing and analyzing ECDRM data to detect damage of wood structures. Unlike traditional data-driven methods, the system framework combines prior knowledge about the problem to be solved during the learning process, especially the system physics as a constraint in the loss function, to guide the learning process and improve the model's interpretability. In addition, the present application realizes a 1D-CNN containing two convolutional layers and two fully connected layers to learn complex features and patterns from ECDRM data. Integrating physical information into the framework helps to reduce the size of the neural network and speed up the convergence process.

[0006] To achieve the above-mentioned purpose, the present application provides a wood structure damage detection and evaluation method based on a physical information neural network, which comprises:

[0007] collecting ECDRM data of the surface of the wood structure using a data acquisition device;

[0008] inputting the preprocessed ECDRM data into a physical information neural network model to obtain the identification results of the damage type and degree of the wood structure;

[0009] displaying the identification results of the damage type and degree through a graphical interface.

[0010] A further improvement of the present application is that in the preprocessing process of the ECDRM data, a Savitzky-Colay filter is used to process the original ECDRM data.

[0011] A further improvement of the present application is that the physical information neural network model comprises a one-dimensional convolutional neural network (1D-CNN) and a physics-enhanced loss network in series, wherein:

[0012] The one-dimensional convolutional neural network comprises, in sequence from input to output: a convolutional layer Conv1, a batch normalization layer BN1, a leaky rectified linear unit LeakyReLU, a max pooling layer ML1, a convolutional layer Conv2, a batch normalization layer BN2, a leaky rectified linear unit LeakyReLU, and a max pooling layer ML2.

[0013] The physical information enhancement loss network processes the output features of the one-dimensional convolutional neural network, and sequentially includes a full connection layer FC1, a batch normalization layer BN3, a leaky rectified linear unit LeakyReLU, a dropout regularization layer dropout, and a full connection layer FC2 from input to output.

[0014] The output of the physical information enhancement loss network is processed by a softmax function to obtain a recognition result y-pred.

[0015] Further improvement of the present application is that in the process of training the physical information neural network model, the loss function used is: Loss CE +Loss PE ; wherein, Loss CE is a cross-entropy loss, Loos PE is a physical information constraint loss, defined as:

[0016]

[0017] In the formula, N is the number of nodes, V i is the constraint loss value, defined as:

[0018]

[0019] Loss CE is a cross-entropy loss, defined as

[0020]

[0021] In the formula, y i represents the label of sample i, and the positive class is 1 and the negative class is 0; p i represents the probability that sample i is predicted as a positive class.

[0022] Further improvement of the present application is that the construction process of the data set used to train the physical information neural network model includes the following steps:

[0023] (1) Collect ECDRM data: apply ECDRM technology on wood structure, realize accurate detection of internal damage of wood structure by adjusting drilling speed and force; the data acquisition module records the resistance value of each test through drilling experiment, and constructs an ECDRM data set containing multiple positive samples and multiple negative samples; the positive sample is the detection data of damaged wood structure, and the negative sample is the detection data of undamaged wood structure;

[0024] (2) Data enhancement: the minimum value selection enhancement method is used to expand the initially obtained ECDRM data set;

[0025] (3) Adopting Savitzky-Golay filter to filter each piece of the amplified ECDRM dataset.

[0026] A further improvement of the present application is that the Adam optimization algorithm is used in the process of training the physical information neural network model to improve the convergence speed.

[0027] A further improvement of the present application is that in the process of displaying the recognition results of damage types and degrees, the detection results of different positions can be displayed simultaneously through heat maps and column charts.

[0028] To evaluate the performance of the proposed system framework, we used a set of ECDRM measurement data collected from wooden samples with different damage degrees. Through a series of field tests, we obtained a total of 450 independent data samples. Then, by selecting the minimum value of each pair of samples, the dataset was expanded to obtain a total of 23695 samples. Next, the expanded dataset was divided into training and test sets, and the proposed framework was trained and tested on these data. The results show that our framework achieved a high accuracy of more than 95% in detecting damage in wooden samples, outperforming traditional detection methods. In addition, incorporating physical constraints into the loss function helps improve the model's interpretability, enabling us to better understand the relevant factors of damage detection.

[0029] Compared with traditional methods, the intelligent wood structure damage detection and evaluation system based on physical equation constraints has the following advantages:

[0030] 1. Higher accuracy: The system combines the machine learning framework considering physical equation constraints (PINN) and one-dimensional convolutional neural network (1D-CNN), which can extract more complex features and patterns from ECDRM data, achieving higher accuracy in damage detection.

[0031] 2. Stronger interpretability: Incorporating physical constraints into the loss function helps improve the model's interpretability. This enables engineers and practitioners to better understand the relevant factors of damage detection, thus more accurately evaluating the integrity of wooden structures.

[0032] 3. High efficiency: Due to the small size of the model, the system framework converges quickly during the learning process, thus completing training and evaluation in a relatively short time.

[0033] 4. Non-destructive detection: The system uses ECDRM data for damage detection, and ECDRM technology as a semi-destructive detection method has relatively small damage to the wood, which does not affect the structural integrity of the material.

[0034] 5. Automation and standardization: The system framework can automatically analyze and interpret ECDRM data, reducing human interference and errors, and achieving a more standardized detection process.

[0035] Suitable for complex environments: The system framework can handle complex situations in practical applications, such as considering various variables present in the test environment, such as nodes, cracks, or other defects.

[0036] The concept, specific structure and technical effects of the present application will be further described in conjunction with the accompanying drawings to fully understand the purpose, features and effects of the present application. BRIEF DESCRIPTION OF DRAWINGS

[0037] Figure 1 is a schematic diagram of the implementation process of the wood structure damage detection and evaluation method based on the physical information neural network of the present application;

[0038] Figure 2 is a schematic diagram of the ECDRM data acquisition device adopted by the present application;

[0039] Figure 3 is a schematic diagram of ECDRM data before and after data enhancement;

[0040] Figure 4 is a schematic diagram of ECDRM data before and after Savitzky-Golay filtering;

[0041] Figure 5 is a schematic diagram of the physical information neural network model;

[0042] Figure 6 is a schematic diagram of the difference calculation when calculating the physical information enhancement loss function. DETAILED DESCRIPTION

[0043] The following describes the embodiments of the present application through specific examples. Those skilled in the art can easily understand other advantages and effects of the present application from the disclosure in the specification. The present application can also be implemented or applied in different specific embodiments, and the details in the specification can be modified or changed based on different views and applications without departing from the spirit of the present application. It should be noted that the following examples and features in the examples can be combined with each other without conflict.

[0044] It should be noted that the diagrams provided in the following examples only illustrate the basic concept of the present application in a schematic manner, and only show the components related to the present application in the diagrams, not the number, shape and size of the components when actually implemented. The actual implementation of each component may be arbitrarily changed in type, number and proportion, and the layout pattern of the components may also be more complex.

[0045] Some example embodiments of the present application are described for the purpose of illustration, it being understood that the present application can be implemented in other ways not specifically shown in the drawings.

[0046] As shown in Figure 1 , the implementation process of the wood structure damage detection and evaluation method based on physical information neural network of the present application includes the following steps:

[0047] Collecting ECDRM data: As shown in Figure 2 , the present application applies electronic controlled drilling resistance measurement (ECDRM) technology on wood structure, by adjusting the drilling speed and force, to realize the accurate detection of internal damage of wood structure. The data collection module records the resistance value of each test through drilling experiment, and constructs an ECDRM dataset containing 450 samples including 165 positive samples and 285 negative samples. The dataset is classified by binary classification, and the samples are divided into two categories: damaged (positive) and undamaged (negative). The ECDRM curve shows different patterns on the vertical and horizontal axes, reflecting the anisotropic properties of wood.

[0048] Data augmentation: As shown in Figure 3 , in order to improve the quality and diversity of the dataset, the present application uses the minimum selection enhancement method to process the initially obtained 450 samples, generating a dataset of 23,695 samples, which is used to train and evaluate the framework of wood structure damage detection.

[0049] Data preprocessing: As shown in Figure 4 , the present application uses Savitzky-Golay filter for data preprocessing to improve the quality of data and reduce interference. The preprocessing step helps to improve the accuracy of subsequent analysis and eliminate noise and artifacts that may affect wood structure damage detection. Thus, the dataset used to train the physical information neural network model can be obtained.

[0050] Constructing physical information neural network model (PEML-CNN deep learning framework): The present application proposes an innovative model PEML-CNN based on one-dimensional convolutional neural network, which is used to process and classify ECDRM signals. This model introduces batch normalization, L2 regularization and dropout layer technology, so that the model can follow the constraints of physical principles during training, and improve the generalization ability of the model. Its structure is shown in Figure 5 .

[0051] As shown in Figure 5 , Figure 6 , the physical information enhanced loss function Loos CE +Loos PEThe present application proposes a loss function based on physical information enhancement, which combines the principles of physics with the PINN-CNN framework, significantly improving the performance and stability of the wood structure damage detection model. The physical information enhancement loss function incorporates borehole resistance measurement into the calculation, so that the learned feature representation meets the physical constraints of the problem.

[0052] Training and testing: The collected ECDRM dataset is used for training and testing to evaluate the performance of the system framework in damage detection. To ensure the stability and reliability of the model, the dataset will be divided into training set, validation set and test set for multiple cross-validation. During the training process, the model parameters are optimized according to the loss function, using learning rate scheduling and Adam optimization algorithm to improve the damage detection performance, speed up the training process and improve the convergence speed.

[0053] Result analysis: According to the output results of the model, the damage degree of the wood structure is evaluated. By using the interpretability of the model, the key factors leading to damage detection can be further analyzed, providing useful reference information for practical application. In addition, by comparing the performance of the present application with other damage detection methods (such as traditional CNN model), the superiority of the system framework in the field of wood structure damage detection is proved.

[0054] Visualization and reporting: In order to help users better understand the detection results, the damage degree can be displayed in a visual form, such as through heat map, column chart, etc. to represent the distribution and size of the damage area. In addition, detailed damage detection report is generated for professional reference and further analysis.

[0055] Model optimization and update: With the collection and arrangement of more ECDRM data, the system framework can be continuously optimized and updated to adapt to different types and conditions of wood structures.

[0056] Online monitoring and early warning: In practical application, in order to ensure the safety and reliability of wood structure, the present application can be further expanded to real-time online monitoring and early warning function. By deploying multiple ECDRM sensors, the system can provide timely warning information to users when the wood structure is damaged, so as to take necessary maintenance and repair measures.

[0057] Multi-scale damage detection: Considering that wood structures may have multiple scales of damage, the present application also proposes a multi-scale damage detection method. By fusing features of different scales, effective detection of various scale damages can be achieved. This method can improve the accuracy and sensitivity of damage detection, providing more comprehensive and detailed damage information for practical application.

[0058] Damage type identification: In addition to damage degree assessment, the model function can be further expanded to realize the automatic identification of damage type. By collecting ECDRM data of different types of damage, the model is trained to identify and distinguish different types of damage, such as cracks, corrosion, and insect damage. This will help users better understand the causes of damage and provide more reference information for maintenance and repair work.

[0059] In summary, the present application realizes accurate detection and evaluation of wood structure damage by adopting a system framework combining physical information neural network and non-destructive testing technology. Through ECDRM data acquisition, model training and testing, result analysis and visualization, online monitoring and early warning, etc., the overall monitoring of wood structure damage is realized, providing strong support for practical application. At the same time, the present application also has strong scalability and practicality, and can be applied to other types of structure damage detection fields.

[0060] The above examples only illustrate the principles and effects of the present application, and are not intended to limit the present application. Any person skilled in the art can modify or change the above examples without departing from the spirit and scope of the present application. Therefore, all equivalent modifications or changes completed by those skilled in the art without departing from the spirit and technical idea disclosed by the present application should be covered by the claims of the present application.

Claims

1. A method for wood structure damage detection and evaluation based on physical information neural network, comprising: collecting electronic controlled drilling resistance measurement data of a wood structure surface using a data collection device; inputting the preprocessed electronic controlled drilling resistance measurement data into a physical information neural network model to obtain the identification results of the damage type and degree of the wood structure; displaying the identification results of the damage type and degree through a graphical interface; the physical information neural network model comprises a one-dimensional convolutional neural network and a physical information enhanced loss network connected in series, wherein: the one-dimensional convolutional neural network comprises, from input to output, a convolutional layer, a batch normalization layer, a leaky rectified linear unit, a max pooling layer, a convolutional layer, a batch normalization layer, a leaky rectified linear unit, and a max pooling layer; the physical information enhanced loss network processes the output features of the one-dimensional convolutional neural network, and comprises, from input to output, a fully connected layer, a batch normalization layer, a leaky rectified linear unit, a dropout regularization layer, and a fully connected layer; the output of the physical information enhanced loss network is processed by a softmax function to obtain the identification results; In the process of training the physical information neural network model, the loss function adopted is: ; wherein, is the cross-entropy loss, is the physical information constraint loss, defined as: ； wherein is the number of nodes, is the constraint loss value, defined as: ； wherein is the cross-entropy loss defined as: ； wherein the label of the sample is 1 for the positive class and 0 for the negative class; the label of the sample is 1 for the positive class and 0 for the negative class.

2. The method of claim 1, wherein the method is characterized by: in the preprocessing of the electronic controlled drilling resistance measurement data, a Savitzky-Golay filter is used to process the original electronic controlled drilling resistance measurement data.

3. The method of claim 1, wherein the method further comprises: The construction process of the data set used to train the physical information neural network model comprises the following steps: (1) collecting electronic controlled drilling resistance measurement data: applying electronic controlled drilling resistance measurement technology on the wood structure, adjusting the drilling speed and force to realize accurate detection of internal damage of the wood structure; the data collection module records the resistance value of each test through drilling experiments, and constructs an electronic controlled drilling resistance measurement data set containing multiple positive samples and multiple negative samples; the positive samples are the detection data of damaged wood structures, and the negative samples are the detection data of undamaged wood structures; (2) data augmentation: using the minimum value selection augmentation method to augment the initially obtained electronic controlled drilling resistance measurement data set; (3) using a Savitzky-Golay filter to filter each piece of the augmented electronic controlled drilling resistance measurement data set.

4. The method of claim 1, wherein the method further comprises: In the process of training the physical information neural network model, the Adam optimization algorithm is used to improve the convergence speed.

5. The method of claim 1, wherein: In the process of displaying the identification results of the damage type and degree, the detection results of different positions can be displayed simultaneously through a heat map and a column chart.

Images