A kind of vibration signal and random forest algorithm based on four-legged mechanical dog sidewalk void brick detection method

Abstract

The present application relates to the technical field of ground void detection, and especially relates to a kind of footpath void brick detection method based on four-legged mechanical dog vibration signal and random forest algorithm, the method is based on vibration signal acquisition technology obtains the vibration signal generated in the walking process of four-legged mechanical dog on footpath brick pavement, first, signal is continuously wavelet transformed to complete preliminary screening of exception, then signal feature extraction and signal processing method are used to construct vibration information model, finally, random forest algorithm is used for feature distinction, and loose score is calculated, to realize footpath void brick identification detection.Compared with the traditional manual inspection mode, the present application can exclude the influence of subjective factors of inspectors, and has high accuracy;Compared with the existing footpath image recognition technology, the present application can effectively avoid visual misjudgment, has clear physical meaning, is suitable for current footpath void brick detection use, and can effectively improve the accuracy of detection.

Description

Technical Field

[0001] This invention relates to the field of ground void detection technology, and in particular to a method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm. Background Technology

[0002] Voiding in sidewalk paving refers to the appearance of gaps between the bottom of the paving bricks and the base layer under traffic load and environmental influences. Walking on these gaps causes significant shaking and loosening of the paving, increasing the risk of pedestrian falls and other accidents, directly impacting the safety and comfort of pedestrians on sidewalks. Therefore, it is of great importance to scientifically, accurately, and clearly detect voiding in sidewalk paving.

[0003] According to the "Technical Specification for Urban Road Maintenance" (CJJ36-2016), the existing method for inspecting sidewalk paving voids still involves manual measurement using a 10m line. This method lacks standardized precision, is highly susceptible to operator subjectivity, and is inefficient, making continuous inspection of sidewalk paving over large areas difficult. Current technologies lack accurate and efficient detection methods for sidewalk paving voids, and their detection capabilities need improvement. Therefore, this invention provides a sidewalk paving void detection method based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the existing technology and provide a method for detecting loose brickwork in sidewalks based on the vibration signal of a quadrupedal mechanical dog and a random forest algorithm.

[0005] The objective of this invention can be achieved through the following technical solutions:

[0006] This invention provides a method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm, comprising the following steps:

[0007] Step 1: Collect vibration signal data of the quadrupedal mechanical dog during its low-to-medium speed movement using vibration sensors, including vibration displacement, vibration velocity, and vibration frequency along the Z-axis.

[0008] Step 2: Clean the vibration signal data and perform wavelet transform processing to complete the initial screening of anomalies based on the vibration displacement signal characteristics along the Z-axis.

[0009] Step 3: Re-inspect the abnormal area by controlling the robot dog to continuously step on the abnormal area and collect the re-inspection signal;

[0010] Step 4: Clean and group the re-inspection signals, calculate the mean vibration displacement, mean vibration frequency, and root mean square value of vibration velocity for each group; perform power spectral density analysis on the vibration velocity.

[0011] Step 5: Based on the power spectral density analysis results, select the power spectral densities at 2Hz and 4.5Hz as one of the features, and retain the calculated mean vibration displacement, mean vibration frequency, and root mean square vibration velocity; merge the above data into a feature dataset;

[0012] Step 6: Standardize the feature dataset and input it into the trained random forest model for prediction. The output prediction result is the voiding status of the sidewalk brickwork. Based on the output prediction result, construct a loosening score model, call the loosening score model to calculate the score for the voided bricks, and obtain the score result, which is the degree of voiding of the voided brickwork.

[0013] The range of low and medium speeds in step 1 is [1,2], with units of m / s.

[0014] In step 1, the frequency of the vibration signal data collected by the vibration sensor during the low-to-medium speed movement of the quadrupedal mechanical dog shall not be less than 100 Hz, and the return frequency shall not be less than 200 Hz.

[0015] The vibration signal data in step 1 are grouped in seconds.

[0016] The preprocessing in step 2 specifically includes deleting abnormal data based on the sampling frequency, rotation frequency, and the number of data entries per specific second.

[0017] The continuous wavelet in step 2 uses CMOR, and the scale parameter of CMOR is set in the range of (0, 128).

[0018] The anomaly screening in step 2 uses the moving window mean method for identification.

[0019] The standardization transformation in step 6 uses Z-score standardization, which can be achieved by taking the mean and variance standardization. The specific process is as follows:

[0020]

[0021] Where X is the selected Z-axis vibration signal, μ is the mean of the Z-axis vibration signal, and σ is the standard deviation of the Z-axis vibration signal.

[0022] The loosening score model in step 6 is Score = PSD * psd_value + base_value; where psd_value is the shap value corresponding to PSD, and base_value is the expected value of the model.

[0023] The specific situation of the sidewalk brickwork detachment in step 6 includes: classifying the 0-1 categorical variables as the normal and detached situations of the mechanical dog stepping on the brickwork.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] 1. This invention provides a method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm. This method can effectively solve the problems of subjective factors of inspectors, low detection accuracy, and low efficiency in existing manual inspection methods. It can intuitively and accurately detect voids in sidewalk brickwork and reflect the safety and comfort evaluation of the sidewalk.

[0026] 2. This invention controls a mechanical dog to move and obtain vibration signals. The vibration signals are preprocessed, feature indicators are calculated to establish a dataset, the dataset is segmented and standardized, and a random forest model is applied for training, prediction, and evaluation. A loosening score model is established, and the model is used to calculate the loosening score, reflecting the degree of detachment of the brickwork. This invention effectively avoids visual misjudgment, has clear physical meaning, and is suitable for current sidewalk detachment detection, effectively improving detection accuracy.

[0027] 3. In training the random forest model, this invention allows for the setting of a certain number of decision trees, which improves model stability and reduces computational load. When constructing each tree, the random forest randomly selects subsets of samples and features, increasing model diversity and improving the overall model's generalization ability to new data.

[0028] 4. This invention uses the Welch method to estimate the power spectral density of vibration velocity signals, which enhances the utilization of data, has a wide range of applications, and is simple to use. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention;

[0030] Figure 2 A schematic diagram for the initial screening of low-frequency information of Z-axis vibration displacement signal;

[0031] Figure 3 This is a schematic diagram of the mean distribution of vibration displacement along the Z-axis;

[0032] Figure 4 This is a schematic diagram of the mean distribution of vibration frequencies along the Z-axis.

[0033] Figure 5 This is a schematic diagram of the root mean square value distribution of vibration velocity along the Z-axis;

[0034] Figure 6 This is a schematic diagram of the power spectral density distribution of the Z-axis vibration velocity.

[0035] Figure 7 This is a schematic diagram of the ROC curve of the random forest model training results;

[0036] Figure 8This is a schematic diagram illustrating the application of the loosening score model. Detailed Implementation

[0037] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. These embodiments are based on the technical solution of the present invention and provide detailed implementation methods and specific operating procedures. However, the scope of protection of the present invention is not limited to the following embodiments.

[0038] This embodiment provides a method for detecting loose brickwork on sidewalks based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm. This method mainly utilizes vibration signal acquisition, wavelet transform decomposition, vibration signal feature calculation, application of the random forest algorithm, and loosening score model to achieve accurate detection of loose brickwork on sidewalks.

[0039] like Figure 1 As shown, this method can be divided into three stages: each stage can be further subdivided, and the specific steps are as follows:

[0040] Phase 1: Vibration signal acquisition and initial screening

[0041] Vibration signal acquisition technology is a method for accurately measuring and analyzing the vibration characteristics of mechanical equipment using advanced sensors, signal processing, and computer technology. Pedestrian sidewalk paving voids refer to gaps appearing between the bottom of the paving bricks and the base layer under traffic loads and environmental influences. Walking on these gaps causes significant swaying and loosening of the paving bricks, easily leading to pedestrian falls and other accidents, directly affecting the safety and comfort of pedestrians on sidewalks. Vibration signal acquisition technology can obtain vibration signals from a robotic dog moving along a sidewalk. Continuous wavelet transform is then used for initial anomaly screening, followed by signal re-examination. Vibration signal feature processing and a random forest algorithm are used to classify and identify voided paving bricks, and a loosening score is calculated to quantitatively evaluate the extent of the voids.

[0042] First, a vibration sensor device is configured for the robot dog. In this embodiment, the installation location is the top plate on the back of the robot dog. The sensor needs to collect the vibration velocity, vibration displacement and vibration frequency along the Z-axis. The sampling frequency is not less than 100Hz and the feedback frequency is not less than 200Hz. Data cleaning is performed based on the above sampling frequency and abnormal frequency.

[0043] Step 1: Control the robot dog to move on the sidewalk brickwork. Use vibration sensors to acquire vibration signal data during the robot dog's low-speed movement, including vibration velocity, vibration displacement and vibration frequency along the Z-axis.

[0044] Preferred, the selectable range for medium and low speed travel is [1,2] (m / s). After obtaining the vibration signal data returned by the sensor, the data is grouped in seconds. The vibration sensor used has a sampling frequency of 100 and a feedback frequency of 200. Groups with the number of data points within (200,210) are retained.

[0045] Step 2: Preprocess the vibration signal data, identify abnormal second data and clean it. Specifically, the abnormal second data can be deleted based on the sampling frequency, rotation frequency and the number of data points in a specific second.

[0046] Step 3: Perform wavelet transform on the vibration signal data to complete the initial screening of anomalies based on the characteristics of the low-frequency Z-axis vibration displacement signal;

[0047] Specifically, a continuous wavelet transform is performed on the Z-axis vibration velocity signal. The wavelet can be selected as CMOR, and the scale is set to (0, 128). For low-frequency signal anomaly screening, the lowest scale layer is selected, and the corresponding low-frequency Z-axis vibration velocity signal anomaly area is selected. The initial anomaly screening can use the moving window mean method to determine the anomaly location. The threshold is set to 1.2 times the mean amplitude of the main information frequency band. Segments exceeding the threshold are identified as initial anomaly signals. See [link to initial signal, moving standard deviation, and anomaly diagram]. Figure 2 .

[0048] The specific process of the moving window mean method is as follows:

[0049] For the signal X = {x1, x2, ..., x...} n}, set a window size w, and for each time point t, the window mean... for:

[0050]

[0051] The window standard deviation is:

[0052]

[0053] The second stage involves vibration signal re-examination and feature engineering.

[0054] Step 4: Re-inspect the abnormal areas screened out in the previous step. Control the robot dog to continuously step on the abnormal areas and collect re-inspection signals.

[0055] Preferably, the frequency for repeated continuous treading is 2 times / second.

[0056] Step 5: Clean the re-inspection signal data and group it into groups based on seconds. Perform feature calculation and extraction, and calculate the mean vibration displacement, mean vibration frequency, and root mean square (RMS) value of vibration velocity for each group. Perform power spectral density analysis on the Z-axis vibration velocity, selecting the PSD values ​​of the signals at 2Hz and 4.5Hz. Preferably, the RMS values ​​of the signals in each group can be calculated using the RMS formula, and the power spectral density can be estimated using the Welch method. The Welch method can enhance the utilization of data, has a wide range of applications, and is simple to use.

[0057] The specific steps are as follows:

[0058] (1). The calculation of the mean Z-axis vibration displacement and vibration frequency is as follows:

[0059]

[0060]

[0061] Among them, X Zi To collect the Z-axis vibration displacement value at a certain moment, F Zi The Z-axis vibration frequency value is collected at a certain moment.

[0062] (2). Calculate the root mean square of the vibration velocity along the Z-axis. The specific process is as follows:

[0063]

[0064] Where V is the vibration velocity along the Z-axis at a certain moment.

[0065] (3) Power spectral density analysis was performed on the Z-axis vibration velocity, and the Welch method was used to estimate the power spectral density. The Welch method first divides the signal into multiple overlapping or non-overlapping segments, then applies a window function to each segment, and finally applies a Discrete Fourier Transform (DFT) to each segment to obtain the power spectral density estimate for each segment:

[0066]

[0067] Among them, X k (f) represents the Fourier transform result of a certain signal segment.

[0068] By averaging the periodograms across all segments, we obtain an estimate of the power spectral density of the entire signal.

[0069]

[0070] Step 6: Using the obtained power spectral density estimation results, select the power spectral density of the signal at 2Hz and 4.5Hz as one of the features, and retain the calculated mean vibration displacement, mean vibration frequency and root mean square vibration velocity as the feature dataset.

[0071] The characteristic distribution of samples corresponding to normal and void-filled brickwork conditions is as follows: Figures 3 to 6 As shown, Figure 3 As shown, there is a significant difference in the mean vibration displacement along the Z-axis between normal and void-filled brickwork. Figure 4 As shown, there is a significant difference in the mean vibration frequency distribution along the Z-axis between normal and void-filled brickwork. Figure 5 As shown, there is a significant difference in the root mean square distribution of vibration signals along the Z-axis between normal and void-filled brickwork. Figure 6 As shown, there are significant differences in the spectral density values ​​of the two samples at 2 Hz and 4.5 Hz. One-tailed Mann-Whitney U tests were performed on the five characteristics of the two samples, and the p-values ​​were all much less than 0.05, demonstrating that the distributions of the two samples on the five characteristics are statistically significant.

[0072] The third stage involves the random forest model and the loosening score model.

[0073] Step 7: Using the established feature dataset, standardize the data and split the training and test sets. Then, use the random forest algorithm to train the random forest model and save the trained random forest model.

[0074] Preferably, data standardization transformation can be achieved by taking the mean and variance standardization.

[0075] Random forest is an ensemble learning method that, when used as a classification model, utilizes multiple decision trees to classify samples and determines the final category through a "voting" mechanism. When constructing each tree, random forest randomly selects subsets of samples and features to increase model diversity and improve the overall model's ability to generalize to new data.

[0076] Add a 0-1 categorical variable to the dataset to represent normal bricklaying and hollow bricklaying, as the model classification result. Divide the dataset into training and test sets in an 8:2 ratio and perform standardization using Z-score standardization. The specific process is as follows:

[0077]

[0078] Where X is the selected Z-axis vibration signal, μ is the mean of the Z-axis vibration signal, and σ is the standard deviation of the Z-axis vibration signal.

[0079] The randomized spherical model is trained on a partitioned training set, and parameter tuning can be performed using grid search. The parameters are set as follows: 200 decision trees, maximum tree depth 14, and minimum number of samples per split node 4. After training, the model is tested using a test set. The model outputs a 0 or 1 prediction value for each test sample, indicating whether the bricks stepped on by the robotic dog per second have become loose. By calculating metrics such as overall precision, recall, and F1 score on the test set, the model's generalization ability and prediction accuracy can be quantified.

[0080] Step 8: Use the trained random forest model to build a loosening score model.

[0081] Specifically, the loosening score model is constructed based on the results of the random forest model. Preferably, in this embodiment, the SHAPA algorithm is used to obtain the importance of the 2Hz PSD value feature and the basic expectation value of the model in the random forest model results, and the loosening score formula is constructed as follows:

[0082] Score=PSD(2Hz)*psd_value+base_value

[0083] Where psd_value is the shap value corresponding to PSD (2Hz), and base_value is the base expectation value of the model.

[0084] Applying the SHAPA algorithm and considering score normalization, the loosening score model is constructed as follows:

[0085]

[0086] In addition, the SHAPA algorithm can be used to obtain the 4.5Hz PSD value in the random forest model results, and PSD values ​​in other frequency bands can also be obtained.

[0087] The loosening score is calculated using the loosening score model. The score of the loosened brickwork in the re-inspection sample is the degree of loosening of the brickwork.

[0088] Specifically, the model training results show that with 1662 training samples and 336 test samples, the model's accuracy is 91.38%, the AUC area is 0.969, and the cross-validation score is 0.923, indicating good results and demonstrating the model's excellent classification ability. The ROC curve of the example random forest model is shown below. Figure 7 As shown, the points on the curve are closer to the upper left corner, indicating that the model has strong diagnostic performance.

[0089] Step 9: Perform initial screening, re-examination, and standardization transformation on the test sample data to be identified under the same conditions as the training set. Call the saved random forest model for prediction. Call the loosening score model to calculate the loosening score of the detached bricks and output the prediction results: the prediction results of 0-1 variables correspond to normal bricklaying and detached bricklaying, which is the detachment status of the sidewalk bricklaying corresponding to the test sample. Calculate the loosening score of the detached bricks, which corresponds to the degree of detachment of the detached bricklaying.

[0090] Applying the SHAPA algorithm, the loosening score model is constructed as follows:

[0091] Score = PSD(2Hz) * 0.035 + 0.47

[0092] The loosening score model was applied to the test samples of the two types of voids. The comprehensive score for severely voided brickwork was 0.52, and the comprehensive score for moderately voided brickwork was 0.399. See [link to results] for details. Figure 8 The score for loose bricks is related to the degree of loosening.

[0093] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm, characterized in that, Includes the following steps: Step 1: Collect vibration signal data of the quadrupedal mechanical dog during its low-to-medium speed movement using vibration sensors, including vibration displacement, vibration velocity, and vibration frequency along the Z-axis. Step 2: Clean the vibration signal data and perform wavelet transform processing to complete the initial screening of anomalies based on the vibration displacement signal characteristics along the Z-axis. Step 3: Re-inspect the abnormal area by controlling the robot dog to continuously step on the abnormal area and collect the re-inspection signal; Step 4: Clean and group the re-inspection signals, calculate the mean vibration displacement, mean vibration frequency, and root mean square value of vibration velocity for each group; perform power spectral density analysis on the vibration velocity. Step 5: Based on the power spectral density analysis results, select the power spectral densities at 2Hz and 4.5Hz as one of the features, and retain the calculated mean vibration displacement, mean vibration frequency, and root mean square vibration velocity; merge the above data into a feature dataset; Step 6: Standardize and transform the feature dataset, input it into the trained random forest model for prediction, and output the prediction result as the void status of the sidewalk brickwork; construct a loosening score model based on the output prediction result, call the loosening score model to calculate the score for the voided bricks, and obtain the score result, which is the degree of void in the voided brickwork.

2. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 1, characterized in that, The range of low and medium speeds in step 1 is [1,2], with units of m / s.

3. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 2, characterized in that, In step 1, the frequency of the vibration signal data collected by the vibration sensor during the low-to-medium speed movement of the quadrupedal mechanical dog shall not be less than 100 Hz, and the return frequency shall not be less than 200 Hz.

4. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 1, characterized in that, The vibration signal data is grouped in seconds.

5. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 4, characterized in that, The preprocessing in step 2 specifically includes deleting abnormal data based on the sampling frequency, rotation frequency, and the number of data entries per specific second.

6. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 5, characterized in that, The continuous wavelet in step 2 uses CMOR, and the scale parameter of CMOR is set in the range of (0, 128).

7. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 6, characterized in that, The anomaly screening in step 2 uses the moving window mean method for identification.

8. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 7, characterized in that, The standardization transformation in step 6 uses Z-score standardization, which can be achieved by taking the mean and variance standardization. The specific process is as follows: Where X is the selected Z-axis vibration signal, μ is the mean of the Z-axis vibration signal, and σ is the standard deviation of the Z-axis vibration signal.

9. The method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm according to claim 8, characterized in that, The loosening score model in step 6 is Score = PSD * psd_value + base_value; where psd_value is the shap value corresponding to PSD, and base_value is the expected value of the model.

10. A method for detecting voids in sidewalk brickwork based on vibration signals from a quadrupedal mechanical dog and a random forest algorithm, as described in claim 9, is characterized in that... The specific situation of the sidewalk brickwork detachment in step 6 includes: classifying the 0-1 categorical variables as the normal and detached situations of the mechanical dog stepping on the brickwork.

Images