A flat beam spatial behavior analysis method and system

Abstract

This application provides a method and system for analyzing the spatial behavior of flat wire harnesses, comprising: S2, extracting time-series features from a preliminary spatial coordinate distribution, using Fourier transform to separate low-frequency motion trends and high-frequency noise components, and using a filtering algorithm to remove the influence of environmental interference to obtain a smoothed spatial position trajectory; S4, constructing a dynamic attitude model of the wire harness through a continuous displacement change sequence, using 3D reconstruction technology combined with Euler angle transform to extract attitude change parameters from the spatial position trajectory to obtain a real-time attitude update matrix; S7, generating a spatial position prediction model of the wire harness through dynamic feature parameters, using a long short-term memory network to process the continuous data sequence, and outputting the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory. This invention achieves high-precision monitoring, prediction, and anomaly detection of the spatial behavior of flat wire harnesses, providing a reliable basis for the intelligent manufacturing optimization of flat wire harnesses.

Description

Technical Field

[0001] This invention relates to the field of intelligent manufacturing technology, and in particular to a method and system for analyzing the spatial behavior of flat wire harnesses. Background Technology

[0002] Spatial position monitoring and relative displacement analysis of flat wire harnesses are indispensable research directions in modern manufacturing and engineering fields, especially in high-precision applications such as aerospace, automotive manufacturing, and robotics, where their importance is self-evident. With the increasing complexity and integration of equipment, wire harnesses, as core components for connection and transmission, directly affect the stability and reliability of the system through their spatial behavior.

[0003] However, accurately and in real-time monitoring of dynamic characteristics in complex environments has become crucial for driving technological progress. Currently, traditional monitoring methods largely rely on static measurements or single-sensor technologies, such as laser ranging or fixed-point calibration. While these solutions are adequate under ideal conditions, they often exhibit shortcomings in accuracy, adaptability, or real-time performance in dynamic, complex, or multi-interference environments. The limitations of existing methods are particularly evident when the harness is in irregular motion or affected by external factors such as temperature and vibration.

[0004] The core challenges in this field mainly focus on the following technical factors: first, insufficient dynamic capture capability of spatial position; second, difficulty in accurately quantifying relative displacement; and third, interference suppression issues in complex environments. Because spatial position cannot be perceived in real time and comprehensively, it is difficult to accurately model the attitude changes of the harness during movement; simultaneously, the measurement of relative displacement is distorted due to the lack of high-resolution continuous data support; furthermore, environmental interference such as electromagnetic noise or physical obstruction further exacerbates the complexity of monitoring. These unresolved technical factors make it difficult to achieve efficient prediction and control of flat harness behavior in practical applications, leading to unique challenges such as limited system design optimization and difficulties in fault diagnosis.

[0005] Therefore, how to monitor the spatial position of flat wire harnesses in real time and accurately quantify their relative displacement in complex environments has become a key issue in improving system reliability and performance. This research focuses on overcoming the accuracy bottleneck of dynamic capture, improving the resolution of displacement measurement, and suppressing the influence of environmental interference, in order to provide a practical technical path for wire harness behavior analysis under complex working conditions. Summary of the Invention

[0006] To address the problems mentioned in the background art, a first aspect of the present invention provides a method for analyzing the spatial behavior of flat wire harnesses, the method comprising:

[0007] S1, acquire multi-dimensional sensor data of flat wire harness, and collect raw signals of spatial position and motion trajectory in complex environment by deploying distributed laser rangefinder and accelerometer array, generate high-frequency sampling sequence for dynamic capture, and obtain preliminary spatial coordinate distribution;

[0008] S2, extract time series features from the initial spatial coordinate distribution, use Fourier transform to separate low-frequency motion trends and high-frequency noise components, and use filtering algorithms to remove the influence of environmental interference to obtain a smoothed spatial position trajectory;

[0009] S3. Calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, supplement intermediate data points by interpolation. Generate a continuous displacement change sequence for the relative displacement and determine the high-resolution displacement quantization result.

[0010] S4. A dynamic attitude model of the harness is constructed through a continuous sequence of displacement changes. The three-dimensional reconstruction technology combined with Euler angle transformation is used to extract the parameters of attitude change from the spatial position trajectory to obtain the real-time attitude update matrix.

[0011] S5 performs secondary correction for environmental interference on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, Kalman filtering is used to fuse multi-sensor data to suppress the influence of interference on attitude changes and obtain a stable attitude feature sequence.

[0012] S6. Obtain stable attitude feature sequences and high-resolution displacement quantization results, use time series analysis methods to calculate the spatial behavior trend of the harness, extract key motion patterns from attitude changes, and determine the dynamic feature parameters of behavior analysis.

[0013] S7 generates a spatial position prediction model for the harness through dynamic feature parameters, uses a long short-term memory network to process continuous data sequences, and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory.

[0014] S8 extracts anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through the feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results.

[0015] S9 updates the behavior analysis database based on the calibrated monitoring results, extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

[0016] Optionally, step S1 involves acquiring multi-dimensional sensor data from a flat wire harness. This is achieved by deploying a distributed laser rangefinder and accelerometer array to collect raw signals of spatial position and motion trajectory in a complex environment. A high-frequency sampling sequence is generated for dynamic capture to obtain a preliminary spatial coordinate distribution, including:

[0017] Step S11: Collect multi-dimensional sensor data through a distributed laser rangefinder and accelerometer array to generate raw signals;

[0018] Step S12: Obtain the sampling sequence from the original signal using high-frequency sampling technology;

[0019] Step S13: Calculate the spatial coordinates of the sampled sequence and use the least squares method to determine the preliminary spatial coordinates;

[0020] Step S14: If there is a deviation in the spatial coordinates, the coordinate distribution is adjusted by correcting the data from the accelerometer array and using the Kalman filter algorithm.

[0021] Step S15: Based on the adjusted coordinate distribution, extract motion trajectory features using dynamic capture technology;

[0022] Step S16: By comparing the motion trajectory features with a pre-established complex environment database, the trajectory change trend is determined;

[0023] Step S17: Obtain the trajectory change trend, use the support vector machine algorithm to classify the coordinate distribution, and determine the final spatial location distribution based on the classification threshold.

[0024] Optionally, step S12, obtaining a sampling sequence from the original signal using high-frequency sampling technology, includes setting the sampling frequency to 1000Hz.

[0025] Optionally, step S16, which involves comparing the motion trajectory features with a pre-established complex environment database to determine the trajectory change trend, includes:

[0026] Step S161: Collect raw data of motion trajectory using accelerometer and gyroscope, preprocess the sensor data using Kalman filter, and generate initial sequence;

[0027] Step S162: Extract trajectory features based on the initial sequence, and extract frequency domain features using Fast Fourier Transform to obtain a feature set;

[0028] Step S163: Compare and analyze the feature set with the pre-established database, calculate the similarity using Euclidean distance, and determine the matching result;

[0029] Step S164: Determine the trend of change based on the matching results. If the trend deviates from the preset range, adjust the feature weights through a linear regression model.

[0030] Step S165: Update the spatial distribution with the adjusted feature weights and generate the corrected trajectory using particle filtering;

[0031] Step S166: Match the corrected trajectory with the complex environment and use the support vector machine algorithm to determine the final classification;

[0032] Step S167: Obtain the distribution characteristics of the change trend through the classification results to obtain a complete description of the trajectory change.

[0033] Optionally, step S17, which involves obtaining the trajectory change trend, classifying the coordinate distribution using a support vector machine algorithm, and determining the final spatial location distribution based on a classification threshold, includes: the classification threshold being 0.5.

[0034] Optionally, step S3 involves calculating the displacement vector between adjacent time steps based on the smoothed spatial trajectory. If the displacement vector exceeds a preset displacement vector threshold, intermediate data points are supplemented using interpolation to generate a continuous displacement change sequence for the relative displacement, thus determining a high-resolution displacement quantization result. This includes:

[0035] Step S31: Obtain trajectory data of spatial location through Gaussian filtering, calculate displacement vector between adjacent time steps, and obtain preliminary displacement information;

[0036] Step S32: If the displacement vector exceeds the displacement vector threshold, then linear interpolation is used to generate intermediate data points to determine the supplemented displacement point sequence.

[0037] Step S33: Extract relative displacement from the supplemented displacement point sequence to generate a continuous displacement change sequence and obtain a smoothed change trend.

[0038] Step S34: Based on the changing trend, adjust the sequence resolution using linear interpolation to obtain high-resolution displacement sequence data;

[0039] Step S35: For the high-resolution displacement sequence, use discretization to calculate the quantization value of each time step and determine the quantization distribution of the displacement.

[0040] Step S36: Obtain the key points in the quantization distribution, calculate the overall displacement characteristics of the smoothed trajectory using the mean, and obtain the final quantization result;

[0041] Step S37: Extract core change features from the final quantization results and use K-means clustering to determine the displacement pattern of the trajectory.

[0042] Optionally, in step S5, a secondary correction for environmental interference is performed on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical obstruction are detected, multi-sensor data are fused using Kalman filtering to suppress the impact of interference on attitude changes, resulting in a stable attitude feature sequence, including:

[0043] Step S51: Obtain raw time-aligned data through accelerometer, gyroscope and magnetometer, and use sliding window method to eliminate time deviation between sensors;

[0044] Step S52: Perform low-pass filtering on the raw data to remove high-frequency noise and obtain preprocessed sensor data.

[0045] Step S53: Input the preprocessed sensor data into the extended Kalman filter. The state vector contains quaternions and gyroscope bias, and the observation vector contains accelerometer and magnetometer data.

[0046] Step S54: Update the state prediction equation based on the gyroscope angular velocity, update the observation equation using accelerometer and magnetometer data, and output the filtered attitude quaternion sequence.

[0047] Step S55: Calculate the angular velocity variance of the filtered quaternion sequence. When the angular velocity variance exceeds the angular velocity variance threshold, it is determined to be an abnormal fluctuation.

[0048] Step S56: Median filtering is applied to the data in the abnormal interval, with the window width set to 5 sampling points to obtain the smoothed attitude sequence;

[0049] Step S57: Calculate the Euclidean distance between the smoothed sequence and the original filtered sequence. If the Euclidean distance exceeds the Euclidean distance threshold, it is determined that there is residual interference.

[0050] Step S58: Assign weights to each sensor based on its signal-to-noise ratio, and perform weighted fusion of the three-axis data.

[0051] Step S59: Input the weighted data into the complementary filter and dynamically adjust the high-pass and low-pass filter coefficients, wherein the gyroscope data is filtered through the high-pass filter and the accelerometer and magnetometer data are filtered through the low-pass filter.

[0052] Step S510: Output the final stable attitude Euler angles, with an update frequency of 100Hz.

[0053] Optionally, step S55, calculating the angular velocity variance of the filtered quaternion sequence, and determining abnormal fluctuations when the angular velocity variance exceeds the angular velocity variance threshold, includes: the angular velocity variance threshold is 0.1 rad / s.

[0054] Optionally, step S58, which assigns weights to the triaxial data according to the signal-to-noise ratio of each sensor, includes: setting the weight of the accelerometer to 0.6, the weight of the gyroscope to 0.3, and the weight of the magnetometer to 0.1.

[0055] A second aspect of the present invention provides a flat wire harness spatial behavior analysis system, which performs spatial behavior analysis on flat wire harnesses using the method described above, the system comprising:

[0056] The data acquisition module is used to acquire multi-dimensional sensor data of the flat wire harness. By deploying a distributed laser rangefinder and accelerometer array, it collects raw signals of spatial position and motion trajectory in complex environments, generates high-frequency sampling sequences for dynamic capture, and obtains preliminary spatial coordinate distribution.

[0057] The feature extraction module is used to extract time series features from the initial spatial coordinate distribution, use Fourier transform to separate low-frequency motion trends from high-frequency noise components, and use filtering algorithms to remove the influence of environmental interference to obtain a smoothed spatial location trajectory.

[0058] The filtering module is used to calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, the intermediate data points are supplemented by interpolation. A continuous displacement change sequence is generated for the relative displacement to determine the high-resolution displacement quantization result.

[0059] The displacement calculation module is used to construct a dynamic attitude model of the harness through a continuous sequence of displacement changes. It uses 3D reconstruction technology combined with Euler angle transformation to extract the parameters of attitude change from the spatial position trajectory and obtain the real-time attitude update matrix.

[0060] The attitude modeling module is used to perform secondary correction of environmental disturbances on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, the Kalman filter is used to fuse multi-sensor data to suppress the influence of disturbances on attitude changes and obtain a stable attitude feature sequence.

[0061] The interference correction module is used to obtain stable attitude feature sequences and high-resolution displacement quantization results. It uses time series analysis methods to calculate the spatial behavior trend of the harness, extracts key motion patterns from attitude changes, and determines the dynamic feature parameters of behavior analysis.

[0062] The behavior analysis module is used to generate a spatial position prediction model for the harness through dynamic feature parameters. It uses a long short-term memory network to process continuous data sequences and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory.

[0063] The predictive modeling module is used to extract anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through the feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results.

[0064] The anomaly detection module is used to update the behavior analysis database based on the calibrated monitoring results. It extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

[0065] The technical solutions provided by the embodiments of the present invention have the following beneficial effects:

[0066] This invention provides a method and system for analyzing the spatial behavior of flat wire harnesses. By deploying a distributed sensor array to collect multidimensional data of the wire harness, signal processing technology is used to extract spatial position and attitude information. Through time series analysis and three-dimensional reconstruction, a dynamic attitude model of the wire harness is constructed, and environmental interference correction is performed. Combined with a long short-term memory network, real-time prediction and anomaly detection of the wire harness spatial position are achieved.

[0067] By dynamically optimizing the data acquisition strategy through a feedback mechanism, the monitoring accuracy in complex environments has been improved.

[0068] Ultimately, this invention can extract typical behavioral patterns from long-term trends and generate a quantitative feature set that adapts to multiple scenarios, providing an effective solution for the spatial behavior analysis of flat wire harnesses under various working conditions.

[0069] This invention enables high-precision monitoring, prediction, and anomaly detection of the spatial behavior of flat wire harnesses, providing a reliable basis for the intelligent manufacturing optimization of flat wire harnesses, and also providing important support for applications in related fields. Attached Figure Description

[0070] Figure 1 This is a flowchart of a method for analyzing the spatial behavior of flat wire harnesses according to the present invention.

[0071] Figure 2 This is a schematic diagram of the structure of a flat wire harness spatial behavior analysis system according to the present invention. Detailed Implementation

[0072] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0073] like Figure 1 As shown, in a first aspect, the present invention provides a method for analyzing the spatial behavior of flat wire harnesses, the method comprising:

[0074] S1 acquires multi-dimensional sensor data from a flat wire harness. By deploying a distributed laser rangefinder and accelerometer array, it collects raw signals of spatial position and motion trajectory in a complex environment. For dynamic capture, it generates a high-frequency sampling sequence to obtain a preliminary spatial coordinate distribution.

[0075] Optionally, this step also includes:

[0076] Step S11: Collect multi-dimensional sensor data using a distributed laser rangefinder and accelerometer array to generate a raw signal. Step S12: Use high-frequency sampling technology from the raw signal, setting the sampling frequency to 1000Hz, to obtain a sampling sequence. Step S13: Calculate spatial coordinates for the sampling sequence, using the least squares method to determine preliminary spatial coordinates. Step S14: If there is a deviation in the spatial coordinates, correct them using accelerometer array data, employing a Kalman filter algorithm to obtain an adjusted coordinate distribution. Step S15: Based on the adjusted coordinate distribution, extract motion trajectory features using dynamic capture technology. Step S16: Compare the motion trajectory features with a pre-established complex environment database to determine the trajectory change trend. Step S17: Obtain the trajectory change trend, use a support vector machine algorithm to classify the coordinate distribution, setting a classification threshold of 0.5, to determine the final spatial location distribution.

[0077] For example, the process of acquiring multi-dimensional sensor data and generating raw signals through a distributed laser rangefinder and accelerometer array can be understood as using multiple sensors to work together to capture the spatial information of a target.

[0078] For example, in an indoor navigation scenario, laser rangefinders are distributed around the room to measure the distance between the target and each ranging point in real time. Simultaneously, an accelerometer array records changes in the target's acceleration, generating a multi-dimensional data stream containing both distance and acceleration. The advantage of this approach is that it provides high-precision raw signals, laying the foundation for subsequent processing. High-frequency sampling technology is used on the raw signal, setting the sampling frequency to 1000Hz to obtain a sampling sequence, aiming to capture subtle changes in the signal.

[0079] Specifically, assuming the target is a mobile robot, a sampling frequency of 1000Hz means collecting 1000 data points per second, accurately recording the robot's high-frequency vibrations or rapid displacements over a short period. The advantage of this high-frequency sampling is that it avoids signal distortion, especially in dynamic environments, significantly improving data integrity. Calculating the spatial coordinates of the sampled sequence using the least squares method to determine the initial spatial coordinates is a classic optimization approach.

[0080] For example, in robot localization, distance data collected from multiple laser ranging points is combined with geometric relationships to calculate the robot's initial coordinates on a two-dimensional plane. The least squares method finds the optimal solution by minimizing the sum of squared errors, making it suitable for handling uncertainties caused by sensor noise. The result might be an initial estimate such as x = 2.3 meters and y = 1.5 meters. If there are deviations in the spatial coordinates, the data from the accelerometer array is used for correction, and the Kalman filter algorithm is employed to obtain the adjusted coordinate distribution.

[0081] It should be noted that Kalman filtering dynamically adjusts coordinates by fusing acceleration data and laser ranging data.

[0082] For example, when a robot moves, laser ranging may be inaccurate due to occlusion. Accelerometer-recorded acceleration trends can be used to predict position changes; after adjustment, the coordinates might become x = 2.35 meters and y = 1.48 meters. This method effectively reduces errors and improves robustness. Based on the adjusted coordinate distribution, dynamic capture technology is used to extract motion trajectory features, emphasizing the characterization of the target's motion patterns.

[0083] In one possible implementation, the robot moves continuously for 10 seconds, forming a curve at its coordinates. Dynamic capture technology extracts information such as speed changes and corner features, for example, identifying when the robot suddenly turns after moving in a straight line at a speed of 0.5 m / s. This feature extraction provides crucial information for subsequent analysis. By comparing the motion trajectory features with a pre-established database of complex environments, the trend of trajectory changes is determined, relying on pattern matching.

[0084] For example, the database stores trajectory templates for different indoor paths, such as straight trajectories in a corridor or turning trajectories in a room corner. Comparison reveals a high degree of similarity between the robot's trajectory and the "corridor path," indicating that it is moving along the corridor. This comparison helps understand the target's behavioral trends and improves the contextual awareness of localization. Obtaining trajectory change trends and using a support vector machine algorithm to classify the coordinate distribution, setting a classification threshold of 0.5, determines the final spatial location distribution—a method of intelligent decision-making.

[0085] In one embodiment, the robot's trajectory is divided into two categories: "straight ahead" and "turning." The support vector machine calculates a classification score based on the feature vectors. If the score is greater than 0.5, it is determined to be "straight ahead," ultimately confirming that the robot is located in the center of the corridor at coordinates x = 2.4 meters and y = 1.5 meters. This classification method improves the accuracy of position determination and provides reliable support for navigation tasks.

[0086] Optionally, step S16, which involves comparing the motion trajectory features with a pre-established complex environment database to determine the trajectory change trend, further includes:

[0087] Step S161: Collect raw motion trajectory data using an accelerometer and gyroscope, and preprocess the sensor data using Kalman filtering to generate an initial sequence. Step S162: Extract trajectory features from the initial sequence, and extract frequency domain features using Fast Fourier Transform to obtain a feature set. Step S163: Compare and analyze the feature set with a pre-established database, calculate similarity using Euclidean distance, and determine the matching result. Step S164: Determine the trend based on the matching result; if the trend deviates from a preset range, adjust the feature weights using a linear regression model. Step S165: Update the spatial distribution using the adjusted feature weights, and generate a corrected trajectory using particle filtering. Step S166: Match the corrected trajectory with a complex environment, and determine the final classification using a support vector machine algorithm. Step S167: Obtain the distribution characteristics of the trend based on the classification result to obtain a complete description of the trajectory change.

[0088] For example, collecting raw data of motion trajectory using accelerometers and gyroscopes is a common method of motion sensing.

[0089] For example, in an indoor mobile robot scenario, an accelerometer records changes in acceleration as the robot moves forward and backward, while a gyroscope captures subtle adjustments in its rotation angle. Preprocessing this data using Kalman filtering effectively fuses the signals from both types of sensors.

[0090] For example, assuming the robot moves at a speed of 0.2 m / s, the raw data may be noisy due to environmental vibrations. Kalman filtering generates a smooth initial sequence through prediction and update steps, preserving the true trend of the motion. When extracting trajectory features from the initial sequence, a fast Fourier transform is used to convert the time-domain signal into frequency-domain features.

[0091] In one possible implementation, the robot moves along a straight line for 5 seconds. Fast Fourier Transform analysis shows that the dominant frequency is concentrated at 0.5 Hz, indicating smooth motion. The feature set may include the dominant frequency and amplitude, reflecting the stability of the trajectory. By comparing the feature set with a database, Euclidean distance is used to calculate the similarity and determine the matching result.

[0092] Specifically, the database stores the frequency domain features of preset paths, such as a dominant frequency of 0.5Hz for a straight path. Assuming the dominant frequency of the current feature set is 0.48Hz, Euclidean distance calculation shows a similarity of 0.9 with the straight path, indicating a successful match. This comparison method is simple and efficient. The trend is determined based on the matching results; if it deviates from the preset range, the feature weights are adjusted using a linear regression model.

[0093] Preferably, when the robot turns, the dominant frequency may rise to 0.7Hz, deviating from the straight path. Linear regression analysis is used to analyze the relationship between frequency domain features and path type, and weights are reallocated to highlight the importance of turning features, ensuring that subsequent analysis is more realistic. The spatial distribution is updated by adjusting the feature weights, and particle filtering is used to generate the corrected trajectory.

[0094] In one embodiment, if the robot's position estimation is inaccurate after turning, particle filtering corrects the trajectory to x=3.2 meters and y=2.1 meters by sampling a large number of particles and combining acceleration and gyroscope data. This method can dynamically adapt to complex environments. Based on the matching of the corrected trajectory with the complex environment, a support vector machine algorithm is used for final classification.

[0095] For example, the trajectory shows the robot turning a corner. The support vector machine, based on the feature vector classification score of 0.6, classifies it as a "turn." The classification result is clear and reliable. By obtaining the distribution characteristics of the changing trend through the classification result, a complete description of the trajectory change can be obtained.

[0096] Understandably, in a scenario where the robot moves continuously for 10 seconds, the analysis shows that it first moves straight for 5 seconds, then turns for 2 seconds, and then moves straight for 3 seconds again. This distribution characteristic characterizes the changes in speed and direction, providing comprehensive support for navigation tasks. This step-by-step analysis method is logically rigorous, ensuring the completeness of the trajectory description.

[0097] S2 extracts time series features from the initial spatial coordinate distribution, uses Fourier transform to separate low-frequency motion trends from high-frequency noise components, and uses a filtering algorithm to remove the influence of environmental interference to obtain a smoothed spatial position trajectory.

[0098] Optionally, this step also includes:

[0099] Step S21: Obtain time series data using spatial coordinates, and separate frequency components using Fast Fourier Transform (FFT) to obtain low-frequency trends and high-frequency noise. Step S22: Extract motion trend features from the low-frequency trends using a moving average tool to determine long-term variation patterns. Step S23: Apply Kalman filtering to the high-frequency noise to remove environmental interference from the noise components, obtaining preliminary processing results. Step S24: Adjust the Kalman filter parameters based on the preliminary processing results to obtain a smoothed position trajectory. Step S25: Determine the continuity of the motion trend based on the smoothed position trajectory to determine trajectory stability. Step S26: Analyze the periodicity of the time series using trajectory stability analysis to obtain the final feature description. Step S27: Extract key change points from the final feature description using peak detection to generate an optimized representation of the smoothed trajectory.

[0100] For example, acquiring time series data through spatial coordinates and using Fast Fourier Transform to separate frequency components is a common signal processing method.

[0101] Understandably, this method breaks down the raw data into components of different frequencies.

[0102] For example, when a mobile robot moves through an indoor corridor, sensors record continuous spatial coordinate data. A Fast Fourier Transform (FFT) can convert this data into the frequency domain. The low-frequency component reflects the robot's smooth movement trend, such as a uniform speed of 0.3 meters per second, while the high-frequency component may contain noise introduced by uneven ground or sensor jitter, such as tiny fluctuations of tens of times per second. A moving average tool is used to extract motion trend features from the low-frequency trend, focusing on long-term variation patterns.

[0103] Specifically, a moving average smooths data by taking the average value over a period of time.

[0104] For example, suppose the robot moves over 10 seconds, with coordinate data recorded once per second and a sliding window of 3 seconds. Early data may show a slow increase in position, while later data may stabilize, indicating the robot is approaching its destination. This method highlights trends and reduces the interference of short-term fluctuations. Kalman filtering is applied to remove environmental interference for high-frequency noise, aiming to clean up useless components in the data.

[0105] In one possible implementation, as the robot moves, the sensors generate high-frequency noise due to flashing lights. Kalman filtering preserves the true motion signal through prediction and update steps.

[0106] For example, the raw data might contain 10 irregular jumps per second. After processing, these disturbances are reduced, leaving a result that more closely approximates the actual path. Adjusting the Kalman filter parameters based on the preliminary processing results to obtain a smoothed position trajectory is an optimization of the algorithm.

[0107] Preferably, the initial parameters may assume low noise, but if the actual data indicates significant environmental interference, the parameters are adjusted to enhance the filtering effect.

[0108] For example, the robot path changes from a slightly jagged shape to a smooth curve, and the transition between coordinate points is more natural, which helps with subsequent analysis. The continuity of the motion trend is determined based on the smoothed position trajectory, thus confirming trajectory stability and emphasizing path reliability.

[0109] In one embodiment, the robot moves 5 meters in a straight line, with continuous trajectory points and no obvious jumps, indicating stable motion. If the robot slows down due to obstacles, the trajectory remains continuous, confirming stability. This judgment lays the foundation for long-term tracking. By analyzing the periodicity of the trajectory stability time series, the final feature description is obtained, focusing on regular characteristics.

[0110] For example, a robot repeats an acceleration and deceleration pattern every 3 seconds; after analysis, periodic signals are extracted. This description reflects the inherent laws of motion, facilitating behavior prediction. From the final feature description, peak detection tools are used to extract key change points, generating an optimized representation of a smooth trajectory that highlights important moments.

[0111] In one embodiment, the robot trajectory display shows a sudden change in speed from 0.5 m / s to 0 m / s, with peak detection marking this pause. The optimized trajectory representation connects these key points, simplifying the data while preserving key features. This approach improves trajectory readability and provides a clear basis for navigation tasks.

[0112] S3. Calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, supplement intermediate data points through interpolation to generate a continuous displacement change sequence for relative displacement and determine the high-resolution displacement quantization result.

[0113] Optionally, this step also includes:

[0114] Step S31: Obtain trajectory data of spatial location through Gaussian filtering, calculate the displacement vector between adjacent time steps, and obtain preliminary displacement information. Step S32: If the displacement vector exceeds a displacement vector threshold, use linear interpolation to generate intermediate data points and determine the supplemented displacement point sequence. Step S33: Extract relative displacement from the supplemented displacement point sequence to generate a continuous displacement change sequence, obtaining a smoothed change trend. Step S34: Adjust the sequence resolution using linear interpolation based on the change trend to obtain high-resolution displacement sequence data. Step S35: For the high-resolution displacement sequence, use discretization to calculate the quantization value of each time step and determine the quantization distribution of the displacement. Step S36: Obtain key points in the quantization distribution, calculate the overall displacement characteristics of the smoothed trajectory using the mean, and obtain the final quantization result. Step S37: Extract core change features from the final quantization result and use K-means clustering to determine the displacement pattern of the trajectory.

[0115] For example, trajectory data of spatial location can be obtained by Gaussian filtering.

[0116] Understandably, this method utilizes the characteristics of the Gaussian distribution to smooth the data and reduce the impact of sudden changes.

[0117] For example, suppose a robot moves in a corridor, and its sensors record its coordinates once per second, potentially experiencing slight jitter due to environmental factors. Gaussian filtering, through weighted averaging, makes the position data more continuous. This is used when calculating the displacement vector between adjacent time steps.

[0118] Specifically, the direction and magnitude of displacement per second can be obtained by subtracting pairs of smoothed coordinate points.

[0119] For example, when a robot moves from position A to position B, the coordinate difference between two adjacent seconds forms a vector, which initially reflects the movement trend. If the displacement vector exceeds a preset displacement vector threshold, linear interpolation is required to generate intermediate data points.

[0120] In one possible implementation, assuming a displacement vector threshold of 0.5 meters, a displacement reaching 0.8 meters indicates a potential data jump. Linear interpolation refines the sequence by uniformly inserting data between two points, for example, adding a point every 0.2 seconds. This addition improves the accuracy of subsequent analysis. When extracting relative displacements from the added sequence...

[0121] Preferably, the difference between each point and the previous point can be calculated to form a continuous sequence of changes.

[0122] For example, if a robot's displacement per second changes from 0.3 meters to 0.4 meters, the change sequence records these increments, making the trend clearer. Linear interpolation again comes into play when adjusting the sequence resolution based on the changing trend.

[0123] Specifically, if the original data is one point per second, it can be interpolated to one point every 0.5 seconds, and the high-resolution sequence can capture more subtle changes.

[0124] For example, when a robot decelerates, the interpolated sequence shows the displacement gradually changing from 0.3 meters to 0.1 meters, with richer details. This is especially relevant when discretizing and quantizing high-resolution sequences.

[0125] It is understandable that continuous displacements are mapped to discrete intervals.

[0126] For example, each 0.1 meter is used as a quantification unit; a displacement of 0.35 meters is quantified as 3, and the quantification distribution intuitively reflects the pattern of change. After obtaining the key points in the quantification distribution, the overall displacement characteristics are calculated using the mean.

[0127] In one embodiment, the robot moves over 10 seconds, with quantization values ​​ranging from 2 to 4 and a mean of 3, indicating an average displacement of approximately 0.3 meters per step. This characteristic is concise and practical. When extracting core change features from the final quantization results, K-means clustering can classify the displacement patterns.

[0128] For example, clustering revealed two patterns: one was a constant speed movement of about 0.3 meters, and the other was a slow adjustment of about 0.1 meters. This classification helps to understand the robot's behavior patterns and improves the reliability of trajectory analysis.

[0129] It's important to note that each step is interconnected, progressing from smoothing to clustering, gradually refining the data to ultimately form a clear displacement pattern. This method is logically rigorous and progressive, ensuring that the results are both detailed and holistic.

[0130] Preferably, supplementing data points and adjusting resolution can adapt to different scenario requirements, while quantization and clustering highlight key features, facilitating subsequent applications.

[0131] S4 constructs a dynamic attitude model of the harness through a continuous sequence of displacement changes. Using 3D reconstruction technology combined with Euler angle transformation, the parameters of attitude change are extracted from the spatial position trajectory to obtain the real-time attitude update matrix.

[0132] Optionally, this step also includes:

[0133] Step S41: Obtain the displacement change sequence and use a preset acquisition tool to obtain an initial trajectory point set. Step S42: Extract displacement change features from the initial trajectory point set and calculate dynamic attitude parameters using Euler angle transformation. Step S43: Calculate Euler angles based on the displacement change features to obtain a preliminary attitude description. Step S44: For the preliminary attitude description, use 3D reconstruction technology to generate the spatial distribution of the harness model and determine the 3D attitude framework. Step S45: Update the dynamic attitude parameters based on the 3D attitude framework and use matrix operations to obtain the real-time attitude matrix. Step S46: If the deviation between the real-time attitude matrix and the initial trajectory point set exceeds an attitude deviation threshold, recalculate the attitude parameters by adjusting the Euler angle transformation. Step S47: Adjust the Euler angles based on the deviation and recalculate the attitude parameters to obtain a corrected matrix. Step S48: Obtain the corrected matrix and combine it with sequence construction technology to generate a continuous harness model to determine the integrity of the dynamic attitude. Step S49: Update the technology fusion strategy based on the integrity judgment result and use 3D reconstruction to optimize the initial trajectory point set to obtain the final attitude model.

[0134] For example, when acquiring the displacement change sequence, a preset acquisition tool is used to obtain the initial trajectory point set.

[0135] Understandably, such tools typically rely on sensors to capture motion data.

[0136] For example, a robot moves in a straight line indoors, and the sensors record its position coordinates once per second, forming a preliminary set of points.

[0137] For example, suppose the robot moves from the starting point for 10 seconds, and the data acquisition tool records the x and y coordinates per second, with the point set reflecting its approximate path. When extracting displacement change features from the initial trajectory point set, Euler angle transformation is used to calculate dynamic attitude parameters.

[0138] Specifically, displacement variation characteristics can be characterized by the distance and direction difference between points.

[0139] In one possible implementation, the coordinate difference between two consecutive points on a certain path of the robot indicates directional deflection, and Euler angle transformation converts these changes into yaw, pitch and roll angles.

[0140] For example, one second the robot is facing north, the next second it's slightly to the east, and the yaw angle records this turn. Euler angles are calculated based on the displacement change characteristics to obtain a preliminary attitude description.

[0141] Preferably, the changes can be analyzed from the perspective of continuous point concentration.

[0142] In one embodiment, when the robot turns, the pitch angle is adjusted according to the change in ground slope, and the initial posture describes its approximate orientation and tilt state.

[0143] For example, when the slope increases by 2 degrees, Euler angles reflect this dynamic. For the initial attitude description, 3D reconstruction technology is used to generate the spatial distribution of the harness model to determine the 3D attitude framework.

[0144] It should be noted that the wire harness model forms a spatial structure by connecting trajectory points.

[0145] For example, the robot's movement path is reconstructed as a continuous line, and the frame shows its pose changes in three-dimensional space.

[0146] For example, the wiring harness bends at a turn, reflecting steering characteristics. When updating dynamic attitude parameters based on the 3D attitude framework, matrix operations are used to obtain the real-time attitude matrix.

[0147] Specifically, the matrix can integrate Euler angle data to form a complete description.

[0148] In one possible implementation, the robot updates its pose matrix once per second, recording the real-time states of turning and tilting.

[0149] For example, the real-time attitude matrix shows the yaw angle changing from 0 degrees to 5 degrees. If the deviation between the real-time attitude matrix and the initial trajectory point set exceeds a preset attitude deviation threshold, the attitude parameters are recalculated by adjusting the Euler angle transformation.

[0150] Understandably, the deviation could be due to sensor noise.

[0151] For example, assuming the attitude deviation threshold is 3 degrees, if the matrix deflection reaches 5 degrees in a certain instance, the Euler angles need to be readjusted for calibration. The Euler angles are adjusted based on the deviation, and the attitude parameters are recalculated to obtain the corrected matrix.

[0152] Preferably, the angle value can be finely adjusted to reduce error.

[0153] In one embodiment, the yaw angle is corrected from 5 degrees to 2 degrees, and the correction matrix more closely reflects the actual path, improving consistency. The corrected matrix is ​​obtained and combined with sequence construction techniques to generate a continuous harness model to determine the integrity of the dynamic attitude.

[0154] For example, the wire harness model shows that the robot path has no obvious breaks and has high integrity.

[0155] Specifically, continuous models can intuitively reflect movement trends.

[0156] In one possible implementation, the technical fusion strategy is updated by updating the integrity judgment result. When the initial trajectory point set is optimized by 3D reconstruction to obtain the final attitude model, if the integrity is insufficient, the reconstruction density is increased.

[0157] For example, by increasing the set of trajectory points from one per second to one per 0.5 seconds, the final model captures pose details more accurately. This optimization ensures that the model is comprehensive and reliable.

[0158] S5 performs secondary correction for environmental interference on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, Kalman filtering is used to fuse multi-sensor data to suppress the impact of interference on attitude changes and obtain a stable attitude feature sequence.

[0159] Optionally, this step also includes:

[0160] Step S51: Obtain time-aligned raw data from the accelerometer, gyroscope, and magnetometer, and use a sliding window method to eliminate time deviations between sensors. Step S52: Perform low-pass filtering on the raw data to remove high-frequency noise, obtaining preprocessed sensor data. Step S53: Input the preprocessed sensor data into an extended Kalman filter. The state vector includes quaternions and gyroscope bias, and the observation vector includes accelerometer and magnetometer data. Step S54: Update the state prediction equation based on the gyroscope angular velocity, update the observation equation using the accelerometer and magnetometer data, and output the filtered attitude quaternion sequence. Step S55: Calculate the angular velocity variance of the filtered quaternion sequence. When the angular velocity variance exceeds the threshold of 0.1 rad / s, it is considered an abnormal fluctuation. Step S56: Apply median filtering to the data in the abnormal interval, setting the window width to 5 sampling points to obtain a smoothed attitude sequence. Step S57: Calculate the Euclidean distance between the smoothed sequence and the original filtered sequence. If the Euclidean distance exceeds the threshold of 0.5, residual interference is considered to exist. Step S58: Assign weights based on the signal-to-noise ratio of each sensor: accelerometer weight 0.6, gyroscope weight 0.3, and magnetometer weight 0.1. Perform weighted fusion of the three-axis data. Step S59: Input the weighted data into a complementary filter, dynamically adjusting the high-pass and low-pass filter coefficients. Gyroscope data is filtered through the high-pass filter, while accelerometer and magnetometer data are filtered through the low-pass filter. Step S510: Output the final stable attitude Euler angles, updated at a frequency of 100Hz.

[0161] For example, the sliding window method is key when acquiring time-aligned raw data using accelerometers, gyroscopes, and magnetometers.

[0162] For example, suppose a robot collects data every 0.01 seconds. Due to hardware differences, the three sensors may have slight time differences. By setting the sliding window to 0.05 seconds and aligning the data within the window according to timestamps, the bias can be eliminated.

[0163] Understandably, this method relies on high-frequency sampling by the sensor.

[0164] For example, when the robot is stationary, the accelerometer records the direction of gravity, the gyroscope captures the angular velocity, and the magnetometer points north; a windowing method ensures that the three are synchronized. When the raw data is low-pass filtered, high-frequency noise is often caused by vibration.

[0165] Specifically, during robot movement, motor vibrations cause interference. Low-pass filtering retains the 0-10Hz signal while removing noise above 20Hz.

[0166] In one possible implementation, the data is transformed from jagged spikes into smooth curves, making it more suitable for analysis after preprocessing. When the preprocessed data is input into the extended Kalman filter, the state vector is described by quaternions to represent the attitude.

[0167] Preferably, the gyroscope bias drifts over time and needs to be dynamically estimated.

[0168] For example, when a robot turns, a gyroscope records angular velocity, an accelerometer provides a gravity reference, a magnetometer calibrates direction, and a filter fuses the data from these three sources to output a stable quaternion. When updating state predictions based on the gyroscope's angular velocity, the observation equations are corrected using the accelerometer and magnetometer.

[0169] In one embodiment, when the robot accelerates, the accelerometer detects linear changes, the magnetometer ensures consistent orientation, and the filtered quaternion sequence reflects the real-time attitude. When calculating the angular velocity variance of the quaternion sequence, 0.1 rad / s is used as the angular velocity variance threshold to identify anomalies.

[0170] For example, if a robot is suddenly struck by an external force and its angular velocity fluctuates to 0.15 rad / s, it indicates an anomaly.

[0171] It should be noted that analysis of variance can quickly locate problem intervals. When median filtering is applied to outlier intervals, the data is smoothed using a 5-sampling-point window.

[0172] For example, if a segment of data jumps due to noise, median filtering is used to extract the median value, restoring the continuity of the attitude sequence and making it suitable for subsequent processing. When calculating the Euclidean distance between the smoothed sequence and the original sequence, 0.5 is used as the Euclidean distance threshold.

[0173] For example, a distance of 0.7 indicates that residual interference still exists and further optimization is needed.

[0174] Understandably, distance reflects data consistency. When assigning weights based on sensor signal-to-noise ratio, the accelerometer is given a weight of 0.6 due to its high accuracy, followed by the gyroscope at 0.3, while the magnetometer is given a relatively low weight of 0.1 due to its susceptibility to interference.

[0175] Specifically, for indoor robot movement, the magnetometer is affected by metal, and the weighted design improves the robustness of data fusion. When the weighted data is input into the complementary filter, the filter coefficients are dynamically adjusted.

[0176] For example, gyroscope data is filtered through a high-pass filter to capture rapid changes, while accelerometer and magnetometer data are filtered through a low-pass filter to provide long-term stability, resulting in more reliable attitude measurement after fusion. A 100Hz update frequency meets real-time requirements when outputting the final attitude Euler angles.

[0177] In one embodiment, the yaw angle is updated every 0.01 seconds during robot navigation, accurately reflecting the turning direction and ensuring strong consistency between the preceding and following logic.

[0178] S6. Obtain stable attitude feature sequences and high-resolution displacement quantization results, use time series analysis methods to calculate the spatial behavior trend of the harness, extract key motion patterns from attitude changes, and determine the dynamic feature parameters for behavior analysis.

[0179] Optionally, this step also includes:

[0180] Step S61: Collect raw attitude data through sensors and obtain a stable attitude feature sequence using Kalman filtering. Step S62: Extract displacement information from the attitude feature sequence and obtain high-resolution quantized data through cubic spline interpolation. Step S63: Perform time series analysis on the quantized data using an autoregressive integral moving average model to obtain a spatial behavior description of the harness. Step S64: Extract the changing trend from the spatial behavior description and determine key motion patterns using a dynamic time warping algorithm. Step S65: Obtain dynamic feature parameters based on the key motion patterns and determine the significance of the behavior trend. Step S66: Determine abnormal behavior patterns by comparing the dynamic feature parameters with a preset dynamic feature threshold. Step S67: Classify the abnormal behavior patterns using a K-means clustering algorithm to obtain the final behavior analysis results.

[0181] For example, after acquiring raw attitude data through sensors, Kalman filtering is the basis for obtaining a stable attitude feature sequence.

[0182] For example, when robots rely on accelerometers and gyroscopes to collect data, the raw signals often contain noise due to environmental vibrations or hardware delays.

[0183] In one embodiment, Kalman filtering fuses multi-source data through prediction and update steps, gradually smoothing the signal and ultimately outputting a stable attitude sequence. This method is particularly suitable for scenarios with high real-time requirements. When extracting displacement information from the attitude feature sequence, cubic spline interpolation can significantly improve data resolution.

[0184] For example, the robot records posture data every 0.02 seconds, and after interpolation, displacement points can be generated every 0.005 seconds.

[0185] Specifically, the interpolation process uses a smooth curve between adjacent points to fill the gaps, ensuring continuous displacement changes.

[0186] Preferably, this method is suitable for scenarios requiring high-precision trajectory analysis. For quantified data, the autoregressive integral moving average model is often used to analyze the spatial behavior of time series.

[0187] Understandably, this model predicts future trends using historical data while also taking into account random disturbances.

[0188] For example, when a robot moves along a straight line, the model extracts periodic features from the displacement sequence to describe its spatial behavior.

[0189] It should be noted that the model parameters need to be dynamically adjusted according to the data characteristics to adapt to different motion states. When extracting trends from spatial behavior descriptions, the dynamic time warping algorithm can effectively identify key motion patterns.

[0190] In one possible implementation, the robot performs a turning motion, and the algorithm identifies pattern similarities by comparing displacement sequences at different time points.

[0191] For example, the displacement curve during turning differs significantly from other time periods, and the algorithm identifies this as a key pattern. This method is particularly effective for capturing irregular movements. After obtaining dynamic feature parameters based on key movement patterns, determining the salience of behavioral trends is the next key step.

[0192] Specifically, parameters such as the rate of change of velocity or the angle of directional deviation can reflect behavioral characteristics.

[0193] For example, when the robot's speed abruptly changes from 1 m / s to 0.5 m / s, with a deviation angle exceeding 10 degrees, it indicates a significant trend.

[0194] In one embodiment, this analysis helps to quickly locate the origin of an anomaly. By comparing dynamic feature parameters with preset dynamic feature thresholds, abnormal behavior patterns can be efficiently determined.

[0195] For example, the threshold for the rate of change of velocity is set to 0.3 m / s², and anything exceeding this value is considered abnormal.

[0196] For example, when a robot is moving on a flat surface, it may encounter an obstacle that causes a sudden drop in speed, triggering an anomaly detection.

[0197] Understandably, the design of dynamic feature thresholds needs to be optimized in conjunction with actual scenarios to reduce false positives. When using the K-means clustering algorithm to classify abnormal behavior patterns, similar anomalies can be grouped into one category.

[0198] In one embodiment, the robot slows down multiple times due to obstacles, and the clustering algorithm classifies it into two categories, "slight obstruction" and "severe stagnation," based on changes in speed and direction.

[0199] Specifically, the clustering process relies on the similarity calculation of feature vectors, and the output directly reflects the behavior type. This classification provides a clear basis for subsequent decision-making.

[0200] Optionally, step S65, which involves obtaining dynamic feature parameters based on key motion patterns and determining the significance of behavioral trends, further includes:

[0201] Step S651: Collect raw data from a triaxial accelerometer using a sensor, and process the raw data using a moving average filter to obtain a denoised motion sequence. Step S652: Extract X-axis acceleration data from the motion sequence as a time series, and calculate the power spectral density of the time series using a fast Fourier transform. Step S653: Detect significant peaks in the power spectral density whose amplitude exceeds a preset multiple of the baseline noise, and determine the period value corresponding to the significant peak. Step S654: If the period value exceeds a period threshold, cluster the abnormal time periods in the time series using a K-means algorithm to obtain multiple motion patterns. Step S655: Extract the root mean square value from the motion patterns as dynamic parameters, and calculate the sliding variance of the dynamic parameters. Step S656: If the sliding variance remains below a sliding variance threshold, determine that the dynamic parameters exhibit a steady-state trend, and use a two-sample t-test to calculate the significant difference between the steady-state trend and historical data. Step S657: Obtain anomaly markers and trend lines from the significant differences, and use a drawing tool to draw a time series diagram containing the anomaly markers and trend lines to obtain a behavior display diagram.

[0202] For example, acquiring raw data from triaxial accelerometers via sensors is the foundation of behavioral analysis.

[0203] For example, in monitoring the operation of a robotic arm on a production line, sensors can be installed at key joints of the robotic arm to record acceleration data along the X, Y, and Z axes in real time, with a sampling frequency of 100Hz to capture subtle changes in motion. The raw data often contains noise; using a moving average filter can effectively smooth out these fluctuations.

[0204] Specifically, a sliding window of length 5 is set, and the average value is taken for every 5 consecutive data points to obtain the denoised motion sequence. This preserves the motion trend while reducing errors caused by random interference. The X-axis acceleration data is extracted from the motion sequence as a time series, reflecting the motion characteristics of the robotic arm in a specific direction.

[0205] For example, the X-axis might correspond to the horizontal direction of movement; analyzing its changes can determine whether the robotic arm is operating smoothly. A Fast Fourier Transform is used to calculate the power spectral density, converting the time-domain signal into a frequency-domain signal.

[0206] For example, suppose we acquire X-axis data for 10 seconds, which, after transformation, yields a power spectrum with a frequency range of 0-50Hz. A significant peak appears at 2Hz with an amplitude of 0.8, far exceeding twice the baseline noise threshold. This indicates that the robotic arm exhibits periodic vibration with a period of 0.5 seconds. Detecting significant peaks in the power spectral density and determining the period value is crucial for identifying motion patterns.

[0207] In one possible implementation, if the period value exceeds a preset period threshold of 0.3 seconds, the motion is considered to have abnormal characteristics. Then, the K-means algorithm is used to cluster the data during the abnormal period.

[0208] For example, segments corresponding to 2Hz vibrations in a time series can be extracted and clustered into three categories, representing slight jitter, moderate oscillation, and severe oscillation, respectively. This classification helps in a more detailed analysis of the source of anomalies. Extracting the root mean square value from the motion pattern as a dynamic parameter allows for the quantification of motion intensity.

[0209] For example, the root mean square value for slight fluctuations might be 0.2. Ascending's calculation of the sliding variance reflects the stability of the dynamic parameters.

[0210] Preferably, if the sliding variance remains below a preset sliding variance threshold, such as 0.05, it is considered a steady-state trend. A two-sample t-test is then used to calculate the significant difference from historical data.

[0211] Understandably, the historical data may be a normal operating sequence from a week ago. The test results show a significance P-value of 0.03, which is less than 0.05, indicating that there is a significant difference between the current steady state and the historical state.

[0212] It should be noted that the time series plots with anomaly markers and trend lines visually demonstrate behavioral changes.

[0213] In one embodiment, the plotting tool can mark abnormal periods in red and represent trend lines as smooth blue curves.

[0214] For example, the graph shows the process of gradually returning to stability after an abnormal oscillation. This visualization helps operators quickly identify problems and take corrective action.

[0215] S7 generates a spatial position prediction model for the harness through dynamic feature parameters, uses a long short-term memory network to process continuous data sequences, and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory.

[0216] Optionally, this step also includes:

[0217] Step S71: Acquire dynamic feature data through sensors, and train the continuous data sequence using a Long Short-Term Memory (LSTM) network to obtain a preliminary prediction model. Step S72: Extract parameter generation rules from the preliminary prediction model, and determine the spatial position change trend based on these rules during real-time monitoring. Step S73: Calculate the displacement estimate for the next time step using a Kalman filter based on the spatial position change trend, obtaining the displacement distribution characteristics. Step S74: Determine the attitude adjustment parameters by using the least squares method to judge the offset of the attitude estimate based on the correlation between the displacement distribution characteristics and continuous data. Step S75: Determine the predicted path of the behavior trajectory using linear regression based on the linear relationship between the attitude adjustment parameters and the time step. Step S76: Obtain the path deviation from the predicted path. If the deviation exceeds the path deviation threshold, update the model parameters using gradient descent through the LSM network to obtain the optimized prediction result. Step S77: Output the displacement and attitude estimates for the next time step based on the optimized prediction result, obtaining the final behavior trajectory.

[0218] For example, after acquiring dynamic feature data through sensors, using a long short-term memory network to train a continuous data sequence is a common time series processing method.

[0219] For example, the robot collects motion data every 0.1 seconds using an accelerometer, and the data includes velocity and direction information. Long Short-Term Memory (LSTM) networks can remember data dependencies over a longer period of time, such as the robot's motion trend over 5 consecutive seconds. Through repeated training, the preliminary prediction model can capture the periodic changes in the data.

[0220] Understandably, this method is suitable for processing highly continuous data sequences. When extracting parameters from the initial prediction model to generate rules, the rules can be based on velocity changes or directional shifts.

[0221] For example, if the model identifies a decrease in speed of 0.2 meters per second as a sign of deceleration, it can be used in real-time monitoring to determine whether the spatial position deviates from expectations.

[0222] In one possible implementation, the Kalman filter is used to calculate the displacement estimate for the next time step based on the trend of spatial position change.

[0223] Specifically, if the robot's current displacement is 10 meters and its speed is 1 meter per second, the Kalman filter, combining the prediction from the previous moment with the current sensor data, estimates that the displacement in the next second might be 11 meters. This estimate can also reflect the smoothness of the displacement distribution.

[0224] Preferably, the least squares method can determine the offset of the attitude estimate by correlating the displacement distribution characteristics with continuous data.

[0225] For example, if the robot's actual posture angle deviates from the predicted value by 5 degrees, the least squares method can be used to calculate an adjustment parameter of 3 degrees by fitting historical data, and gradually correct the deviation.

[0226] It should be noted that when the attitude adjustment parameters are used in a linear relationship with the time step, linear regression can generate a predicted path for the behavior trajectory.

[0227] For example, 10 consecutive seconds of robot posture data show that the orientation deviation angle increases by 1 degree per second. Linear regression predicts that the path will deviate from a straight line by about 5 meters in the next 5 seconds. After obtaining the path deviation from the predicted path, if the deviation exceeds a preset path deviation threshold, such as 3 meters, the model needs to be updated.

[0228] In one embodiment, the Long Short-Term Memory network adjusts its parameters using gradient descent, for example, by setting the learning rate to 0.01. After retraining, the bias is reduced to within 1 meter, and the optimized prediction results are closer to reality.

[0229] Specifically, the optimized prediction results output displacement and attitude estimates for the next time step, resulting in a more accurate behavioral trajectory.

[0230] For example, if the robot predicts a displacement of 12 meters in the next second and adjusts its attitude angle to 2 degrees, this result can provide a reliable basis for subsequent navigation.

[0231] Understandably, this method offers greater real-time performance in dynamic environments. From multiple perspectives, sensor data provides the foundation for the model, Kalman filter smooths the estimation, least squares corrects biases, linear regression predicts the path, and gradient descent optimizes the model. These components support each other, ensuring the continuity and accuracy of trajectory prediction. This completeness is particularly crucial for robot motion analysis.

[0232] S8 extracts anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through a feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results.

[0233] Optionally, this step also includes:

[0234] Step S81: Model the behavior trajectory data using the ARIMA algorithm and output an initial predicted value sequence. Step S82: Calculate the absolute deviation value at each time point from the initial predicted value sequence and the actual data sequence. Step S83: Set the preset absolute deviation threshold to 1.5 times the standard deviation of historical deviation data. When the absolute deviation value exceeds the threshold, record the current deviation value and trigger an adjustment rule. Step S84: The adjustment rule sets the sensor sampling frequency to the original frequency multiplied by the ratio of the current deviation value to the absolute deviation threshold, and outputs the adjusted sampling parameter set. Step S85: Perform data acquisition using the adjusted sampling parameter set to generate an optimized actual data sequence. Step S86: Extract the root mean square error (RMSE) from the optimized actual data sequence as the accuracy bottleneck indicator. When this indicator exceeds a preset error upper limit, use the gradient descent algorithm to minimize the RMSE function and output a calibrated sampling interval parameter set. Step S87: Reacquire behavior trajectory data using the calibrated sampling interval parameter set to generate an updated predicted value sequence. Step S88: Calculate the mean deviation between the updated predicted value sequence and the actual data sequence. When the mean deviation is less than 0.8 times the historical mean deviation, the monitoring result is determined to be stable, and the final predicted data sequence is output.

[0235] For example, when modeling behavioral trajectory data using the ARIMA algorithm, it can be understood as a method that combines time series-based autoregression and moving average.

[0236] For example, the robot records position data once per second for 60 seconds, forming a time series containing velocity and displacement. ARIMA analyzes the trends and periodicity of the data to output an initial sequence of predicted values, such as predicting that the displacement in the next 10 seconds might be an increasing sequence of 10 meters, 10.5 meters, 11 meters, etc. This method is suitable for capturing stationary changes in data.

[0237] In one possible implementation, when calculating the absolute deviation from the initial predicted value sequence to the actual data sequence, the difference at each time point can be compared one by one.

[0238] Specifically, assuming the predicted displacement at the 5th second is 10 meters, and the actual displacement is 10.2 meters, the absolute deviation is 0.2 meters. The deviations at 60 time points are calculated sequentially to obtain a deviation sequence. This deviation reflects the gap between the prediction and reality, providing a basis for subsequent adjustments.

[0239] It should be noted that the preset absolute deviation threshold is set to 1.5 times the standard deviation of historical deviation data, which can dynamically adapt to data fluctuations.

[0240] For example, if the historical standard deviation is 0.1 meters, the absolute deviation threshold is 0.15 meters. If the absolute deviation reaches 0.2 meters at a certain moment, exceeding the absolute deviation threshold of 0.15 meters, it is recorded and an adjustment rule is triggered. This mechanism can detect abnormal deviations in a timely manner.

[0241] In one embodiment, the adjustment rule adjusts the sensor sampling frequency by the ratio of the current deviation value to an absolute deviation value threshold.

[0242] Preferably, if the current deviation value of 0.2 meters divided by the absolute deviation threshold of 0.15 meters yields a ratio of approximately 1.33, and the original sampling frequency was 1 time / second, it is adjusted to 1.33 times / second. The new set of sampling parameters increases the data acquisition density, which helps to capture more subtle changes.

[0243] For example, after collecting data using adjusted sampling parameters, an optimized actual data sequence is generated. The robot originally collected data once per second; after adjustment, it collects data once every 0.75 seconds, increasing the number of data points from 60 to 80 within 60 seconds. The new sequence is denser and more realistically reflects the motion trajectory.

[0244] Specifically, when extracting the root mean square error as the accuracy bottleneck indicator from the optimized actual data sequence, it is assumed that the calculated error is 0.05 meters. If the preset upper limit of error is 0.03 meters, the sampling interval is optimized using the gradient descent algorithm if the upper limit is exceeded.

[0245] For example, by iteratively adjusting the sampling interval from 0.75 seconds to 0.7 seconds, the error was reduced to within 0.02 meters. The calibrated parameter set improved data accuracy.

[0246] In one possible implementation, after re-collecting the behavioral trajectory data, an updated sequence of predicted values ​​is generated.

[0247] For example, the robot predicts displacements of 10 meters, 10.4 meters, and 10.8 meters in the next 5 seconds, which are closer to the actual data. The average deviation has decreased from 0.1 meters to 0.06 meters, which is 0.8 times lower than the historical average deviation of 0.08 meters, i.e., 0.064 meters, indicating that the results are stable.

[0248] Understandably, calculating the mean deviation and determining that the monitoring results are stable will result in a more reliable final predicted data sequence.

[0249] For example, the robot's trajectory prediction for the next 10 seconds can be a series of smoothly increasing displacement values. This method, through stepwise optimization, ensures the continuity and accuracy of the prediction, providing solid support for real-time navigation.

[0250] S9 updates the behavior analysis database based on the calibrated monitoring results, extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

[0251] Optionally, this step also includes:

[0252] Step S91: Update the behavior analysis database with calibrated monitoring data to obtain a record set of spatial positions and relative displacements. Step S92: Extract long-term trends from the record set using the K-means clustering algorithm to obtain a set of typical patterns. Step S93: Analyze the harness behavior under complex operating conditions based on the typical pattern set to determine the distribution of behavioral features. Step S94: Generate a multi-scenario adaptive quantitative feature set from the behavioral feature distribution using principal component analysis to determine the coverage of the feature set. Step S95: If the feature set coverage is insufficient, obtain additional spatial position and relative displacement records from the monitoring data to update the typical pattern set. Step S96: Adjust the quantitative feature set based on the updated typical pattern set to obtain the final multi-scenario adaptive feature set. Step S97: Verify the harness behavior under complex operating conditions using the final feature set to determine the completeness of the behavior analysis database.

[0253] For example, the behavior analysis database is updated with calibrated monitoring data to obtain a set of records of spatial location and relative displacement.

[0254] Understandably, this update is to ensure that the database reflects the latest motion trajectory information.

[0255] For example, when a robot moves in a factory environment, its coordinate data is recorded every second, such as starting at 0 meters and reaching 5 meters after 10 seconds, forming a set of records containing both time and displacement. This set provides the basic data for subsequent analysis. The K-means clustering algorithm is used to extract long-term trends from the record set, resulting in a set of typical patterns.

[0256] For example, suppose a database stores a robot's movement data for a week, containing thousands of displacement points. K-means clustering can divide this data into several typical categories, such as "straight-line movement pattern" and "turning pattern".

[0257] Specifically, straight-line movement might show a stable displacement of 0.5 meters per second, while turning patterns exhibit larger displacement fluctuations. This method can effectively summarize long-term behavioral patterns. By analyzing typical pattern sets under complex operating conditions, the behavior of the harness can be determined, and the distribution of behavioral characteristics can be identified.

[0258] In one possible implementation, harness behavior refers to the motion characteristics of a robot when it is in a narrow passage or around an obstacle.

[0259] For example, when a robot navigates around an obstacle, its displacement decreases from 0.5 m / s to 0.2 m / s, accompanied by a change in direction. Analyzing this data may reveal that the behavioral feature distribution is concentrated in two aspects: speed reduction and angle adjustment, providing a basis for subsequent quantification. Principal component analysis is used to generate a multi-scene adaptive quantified feature set from the behavioral feature distribution to determine the coverage of the feature set.

[0260] Preferably, principal component analysis can extract key variables, such as the weights of speed changes and direction adjustments. For example, the analysis results show that speed changes account for 60% of behavioral differences, and direction adjustments account for 30%. A quantified feature set is thus formed, and it is checked whether it covers common scenarios in the factory, such as straight roads and curves. If the coverage is insufficient, the feature set needs further expansion. If the feature set coverage is still insufficient, additional spatial location and relative displacement records are obtained from the monitoring data to update the typical pattern set.

[0261] For example, the original pattern set only included straight-line and turning patterns, but the new data shows the existence of a "pause-acceleration" pattern, such as the robot pausing and then starting at 1 m / s. After the update, the typical pattern set is more comprehensive and can adapt to more changing working conditions. The quantized feature set is adjusted based on the updated typical pattern set to obtain the final feature set adapted to multiple scenarios.

[0262] In one embodiment, after adding the "pause-accelerate" mode, the feature set incorporates acceleration parameters, such as the rate of change from 0 m / s to 1 m / s. The new feature set better describes robot behavior, ensuring adaptability in different scenarios. The final feature set is used to verify the harness behavior under complex working conditions, confirming the completeness of the behavior analysis database.

[0263] Specifically, during verification, a robot can be simulated passing through a narrow passage to check whether the feature set can accurately reflect changes in speed and direction.

[0264] For example, the predicted speed decreased to 0.3 m / s, consistent with the actual data, indicating that the database is complete and reliable. This method improves the practicality of behavior analysis and provides support for optimizing robot navigation.

[0265] like Figure 2 As shown, in a second aspect, the present invention provides a flat wire harness spatial behavior analysis system, which performs spatial behavior analysis on flat wire harnesses using the method described above. The system includes:

[0266] The data acquisition module is used to acquire multi-dimensional sensor data of the flat wire harness. By deploying a distributed laser rangefinder and accelerometer array, it collects raw signals of spatial position and motion trajectory in complex environments, generates high-frequency sampling sequences for dynamic capture, and obtains preliminary spatial coordinate distribution.

[0267] The feature extraction module is used to extract time series features from the initial spatial coordinate distribution, use Fourier transform to separate low-frequency motion trends from high-frequency noise components, and use filtering algorithms to remove the influence of environmental interference to obtain a smoothed spatial location trajectory.

[0268] The filtering module is used to calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, the intermediate data points are supplemented by interpolation. A continuous displacement change sequence is generated for the relative displacement to determine the high-resolution displacement quantization result.

[0269] The displacement calculation module is used to construct a dynamic attitude model of the harness through a continuous sequence of displacement changes. It uses 3D reconstruction technology combined with Euler angle transformation to extract the parameters of attitude change from the spatial position trajectory and obtain the real-time attitude update matrix.

[0270] The attitude modeling module is used to perform secondary correction of environmental disturbances on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, the Kalman filter is used to fuse multi-sensor data to suppress the influence of disturbances on attitude changes and obtain a stable attitude feature sequence.

[0271] The interference correction module is used to obtain stable attitude feature sequences and high-resolution displacement quantization results. It uses time series analysis methods to calculate the spatial behavior trend of the harness, extracts key motion patterns from attitude changes, and determines the dynamic feature parameters of behavior analysis.

[0272] The behavior analysis module is used to generate a spatial position prediction model for the harness through dynamic feature parameters. It uses a long short-term memory network to process continuous data sequences and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory.

[0273] The predictive modeling module is used to extract anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through the feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results.

[0274] The anomaly detection module is used to update the behavior analysis database based on the calibrated monitoring results. It extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

[0275] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the concept of this application. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for analyzing the spatial behavior of flat wire harnesses, characterized in that, The method includes: S1, acquire multi-dimensional sensor data of flat wire harness, and collect raw signals of spatial position and motion trajectory in complex environment by deploying distributed laser rangefinder and accelerometer array, generate high-frequency sampling sequence for dynamic capture, and obtain preliminary spatial coordinate distribution; S2, extract time series features from the initial spatial coordinate distribution, use Fourier transform to separate low-frequency motion trends and high-frequency noise components, and use filtering algorithms to remove the influence of environmental interference to obtain a smoothed spatial position trajectory; S3. Calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, supplement intermediate data points by interpolation. Generate a continuous displacement change sequence for the relative displacement and determine the high-resolution displacement quantization result. S4. A dynamic attitude model of the harness is constructed through a continuous sequence of displacement changes. The three-dimensional reconstruction technology combined with Euler angle transformation is used to extract the parameters of attitude change from the spatial position trajectory to obtain the real-time attitude update matrix. S5 performs secondary correction for environmental interference on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, Kalman filtering is used to fuse multi-sensor data to suppress the influence of interference on attitude changes and obtain a stable attitude feature sequence. S6. Obtain stable attitude feature sequences and high-resolution displacement quantization results, use time series analysis methods to calculate the spatial behavior trend of the harness, extract key motion patterns from attitude changes, and determine the dynamic feature parameters of behavior analysis. S7 generates a spatial position prediction model for the harness through dynamic feature parameters, uses a long short-term memory network to process continuous data sequences, and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory. S8 extracts anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through the feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results. S9 updates the behavior analysis database based on the calibrated monitoring results, extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

2. The method according to claim 1, characterized in that, Step S1 involves acquiring multi-dimensional sensor data from a flat wire harness. By deploying a distributed laser rangefinder and accelerometer array, raw signals of spatial position and motion trajectory are collected in a complex environment. A high-frequency sampling sequence is generated for dynamic capture to obtain a preliminary spatial coordinate distribution, including: Step S11: Collect multi-dimensional sensor data through a distributed laser rangefinder and accelerometer array to generate raw signals; Step S12: Obtain the sampling sequence from the original signal using high-frequency sampling technology; Step S13: Calculate the spatial coordinates of the sampled sequence and use the least squares method to determine the preliminary spatial coordinates; Step S14: If there is a deviation in the spatial coordinates, the coordinate distribution is adjusted by correcting the data from the accelerometer array and using the Kalman filter algorithm. Step S15: Based on the adjusted coordinate distribution, extract motion trajectory features using dynamic capture technology; Step S16: By comparing the motion trajectory features with a pre-established complex environment database, the trajectory change trend is determined; Step S17: Obtain the trajectory change trend, use the support vector machine algorithm to classify the coordinate distribution, and determine the final spatial location distribution based on the classification threshold.

3. The method according to claim 2, characterized in that, Step S12, which involves obtaining a sampling sequence from the original signal using high-frequency sampling technology, includes setting the sampling frequency to 1000Hz.

4. The method according to claim 3, characterized in that, Step S16 involves comparing the motion trajectory features with a pre-established complex environment database to determine the trajectory change trend, including: Step S161: Collect raw data of motion trajectory using accelerometer and gyroscope, preprocess the sensor data using Kalman filter, and generate initial sequence; Step S162: Extract trajectory features based on the initial sequence, and extract frequency domain features using Fast Fourier Transform to obtain a feature set; Step S163: Compare and analyze the feature set with the pre-established database, calculate the similarity using Euclidean distance, and determine the matching result; Step S164: Determine the trend of change based on the matching results. If the trend deviates from the preset range, adjust the feature weights through a linear regression model. Step S165: Update the spatial distribution with the adjusted feature weights and generate the corrected trajectory using particle filtering; Step S166: Match the corrected trajectory with the complex environment and use the support vector machine algorithm to determine the final classification; Step S167: Obtain the distribution characteristics of the change trend through the classification results to obtain a complete description of the trajectory change.

5. The method according to claim 4, characterized in that, Step S17 involves obtaining the trajectory change trend, using a support vector machine algorithm to classify the coordinate distribution, and determining the final spatial location distribution based on a classification threshold, including: the classification threshold being 0.

5.

6. The method according to claim 1, characterized in that, Step S3 involves calculating the displacement vector between adjacent time steps based on the smoothed spatial trajectory. If the displacement vector exceeds a preset displacement vector threshold, interpolation is used to supplement intermediate data points, generating a continuous displacement change sequence for the relative displacement, and determining a high-resolution displacement quantization result, including: Step S31: Obtain trajectory data of spatial location through Gaussian filtering, calculate displacement vector between adjacent time steps, and obtain preliminary displacement information; Step S32: If the displacement vector exceeds the displacement vector threshold, then linear interpolation is used to generate intermediate data points to determine the supplemented displacement point sequence. Step S33: Extract relative displacement from the supplemented displacement point sequence to generate a continuous displacement change sequence and obtain a smoothed change trend. Step S34: Based on the changing trend, adjust the sequence resolution using linear interpolation to obtain high-resolution displacement sequence data; Step S35: For the high-resolution displacement sequence, use discretization to calculate the quantization value of each time step and determine the quantization distribution of the displacement. Step S36: Obtain the key points in the quantization distribution, calculate the overall displacement characteristics of the smoothed trajectory using the mean, and obtain the final quantization result; Step S37: Extract core change features from the final quantization results and use K-means clustering to determine the displacement pattern of the trajectory.

7. The method according to claim 1, characterized in that, Step S5 involves performing secondary correction for environmental interference on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical obstruction are detected, multi-sensor data are fused using Kalman filtering to suppress the impact of interference on attitude changes, resulting in a stable attitude feature sequence, including: Step S51: Obtain raw time-aligned data through accelerometer, gyroscope and magnetometer, and use sliding window method to eliminate time deviation between sensors; Step S52: Perform low-pass filtering on the raw data to remove high-frequency noise and obtain preprocessed sensor data. Step S53: Input the preprocessed sensor data into the extended Kalman filter. The state vector contains quaternions and gyroscope bias, and the observation vector contains accelerometer and magnetometer data. Step S54: Update the state prediction equation based on the gyroscope angular velocity, update the observation equation using accelerometer and magnetometer data, and output the filtered attitude quaternion sequence. Step S55: Calculate the angular velocity variance of the filtered quaternion sequence. When the angular velocity variance exceeds the angular velocity variance threshold, it is determined to be an abnormal fluctuation. Step S56: Median filtering is applied to the data in the abnormal interval, with the window width set to 5 sampling points to obtain the smoothed attitude sequence; Step S57: Calculate the Euclidean distance between the smoothed sequence and the original filtered sequence. If the Euclidean distance exceeds the Euclidean distance threshold, it is determined that there is residual interference. Step S58: Assign weights to each sensor based on its signal-to-noise ratio, and perform weighted fusion of the three-axis data. Step S59: Input the weighted data into the complementary filter and dynamically adjust the high-pass and low-pass filter coefficients, wherein the gyroscope data is filtered through the high-pass filter and the accelerometer and magnetometer data are filtered through the low-pass filter. Step S510: Output the final stable attitude Euler angles, with an update frequency of 100Hz.

8. The method according to claim 7, characterized in that, Step S55 involves calculating the angular velocity variance of the filtered quaternion sequence. When the angular velocity variance exceeds the angular velocity variance threshold, it is determined to be an abnormal fluctuation, including: the angular velocity variance threshold is 0.1 rad / s.

9. The method according to claim 8, characterized in that, Step S58 involves weighting the triaxial data according to the signal-to-noise ratio of each sensor, including setting the weight of the accelerometer to 0.6, the weight of the gyroscope to 0.3, and the weight of the magnetometer to 0.

1.

10. A flat wire harness spatial behavior analysis system, characterized in that, Spatial behavior analysis of flat wire harnesses is performed using the method described in any one of claims 1-9, wherein the system comprises: The data acquisition module is used to acquire multi-dimensional sensor data of the flat wire harness. By deploying a distributed laser rangefinder and accelerometer array, it collects raw signals of spatial position and motion trajectory in complex environments, generates high-frequency sampling sequences for dynamic capture, and obtains preliminary spatial coordinate distribution. The feature extraction module is used to extract time series features from the initial spatial coordinate distribution, use Fourier transform to separate low-frequency motion trends from high-frequency noise components, and use filtering algorithms to remove the influence of environmental interference to obtain a smoothed spatial location trajectory. The filtering module is used to calculate the displacement vector between adjacent time steps based on the smoothed spatial position trajectory. If the displacement vector exceeds the preset displacement vector threshold, the intermediate data points are supplemented by interpolation. A continuous displacement change sequence is generated for the relative displacement to determine the high-resolution displacement quantization result. The displacement calculation module is used to construct a dynamic attitude model of the harness through a continuous sequence of displacement changes. It uses 3D reconstruction technology combined with Euler angle transformation to extract the parameters of attitude change from the spatial position trajectory and obtain the real-time attitude update matrix. The attitude modeling module is used to perform secondary correction of environmental disturbances on the real-time attitude update matrix. If abnormal fluctuations in electromagnetic noise or physical occlusion are detected, the Kalman filter is used to fuse multi-sensor data to suppress the influence of disturbances on attitude changes and obtain a stable attitude feature sequence. The interference correction module is used to obtain stable attitude feature sequences and high-resolution displacement quantization results. It uses time series analysis methods to calculate the spatial behavior trend of the harness, extracts key motion patterns from attitude changes, and determines the dynamic feature parameters of behavior analysis. The behavior analysis module is used to generate a spatial position prediction model for the harness through dynamic feature parameters. It uses a long short-term memory network to process continuous data sequences and outputs the displacement and attitude estimates for the next time step in real-time monitoring to obtain the predicted behavior trajectory. The predictive modeling module is used to extract anomaly detection indicators from the predicted behavior trajectory. If the deviation between the predicted value and the actual collected data exceeds the preset trajectory deviation threshold, the sensor sampling frequency is adjusted through the feedback mechanism to dynamically optimize the data acquisition strategy for the accuracy bottleneck and obtain the calibrated monitoring results. The anomaly detection module is used to update the behavior analysis database based on the calibrated monitoring results. It extracts typical patterns from the long-term trends of spatial location and relative displacement through clustering algorithms, and generates a multi-scenario adaptive quantitative feature set for the behavior of the harness under complex working conditions.

Images