Robot health index intelligent sensing and real-time monitoring method and system

Abstract

The invention belongs to the technical field of artificial intelligence, and particularly relates to a robot health index intelligent sensing and real-time monitoring method and system.The robot health index intelligent sensing and real-time monitoring system comprises a data acquisition module, a data processing module, a health assessment module and an intervention module, the data acquisition module is connected with the data processing module, the data processing module is connected with the health assessment module, and the intervention module is connected with the health assessment module. And the health assessment module is connected with the intervention module. According to the method, data of the robot in the task execution process can be collected and processed in real time, performance decline can be found and quantified in time, and possibility is provided for taking intervention measures in time. According to the invention, an artificial intelligence algorithm is adopted to carry out data processing and health assessment, manual monitoring and analysis are not needed, the labor cost is reduced, and the working efficiency is improved. Noise and interference are eliminated through a filtering algorithm and a machine learning algorithm, key features are extracted, health assessment is carried out by adopting a deep learning algorithm, and the accuracy of an assessment result is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a method and system for intelligent perception and real-time monitoring of robot health indicators. Background Art

[0002] With the continuous development of robot technology, robots are increasingly widely used in various fields. However, during the task execution process, robots often encounter various uncertainties and difficulties, such as terrain obstacles, camera occlusion, sensor noise, and unexpected situations. These factors may lead to a decline in robot performance and affect the task execution effect. Therefore, it is necessary to develop a real-time monitoring system that can automatically detect and quantify the decline in robot performance through artificial intelligence. Summary of the Invention

[0003] In view of the above technical problems that robots often encounter various uncertainties and difficulties during the task execution process, which affect the task execution effect, the present invention provides a method and system for intelligent perception and real-time monitoring of robot health indicators.

[0004] To solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0005] A method for intelligent perception and real-time monitoring of robot health indicators includes the following steps:

[0006] S1. Real-time collect various operation data of the robot during the task execution process, including the robot's position information, speed information, acceleration information, power information, sensor data, and the output result of the artificial intelligence system;

[0007] S2. Preprocess and extract features from the collected data. Use a filtering algorithm to preprocess the original data to eliminate noise and interference, and extract key features that can reflect the decline in robot performance;

[0008] S3. Based on the artificial intelligence algorithm and robot health assessment indicators, evaluate the health status of the robot according to the processed data, and generate robot health indicators;

[0009] S4. When it is detected that the decline in robot performance exceeds the preset threshold, automatically or prompt the user to take intervention measures, including adjusting robot parameters, replacing components, or performing remote control.

[0010] The method for real-time collecting various operation data of the robot during the task execution process in S1 is as follows:

[0011] S1.1. Install a variety of sensors on the robot, including inertial sensors, cameras, lidar, and pressure sensors, for collecting key parameters such as the position, speed, acceleration, temperature, and power of the robot;

[0012] S1.2. Transmit the collected data to the data processing center or cloud server via wireless network or wired connection. For wireless transmission, Wi-Fi, Bluetooth or a dedicated radio frequency module is used. For wired transmission, industrial Ethernet is used.

[0013] The method for preprocessing the original data in S2 is as follows:

[0014] S2.1. Filtering algorithm selection: Use Kalman filtering to process non-linear system noise, use low-pass filtering to eliminate high-frequency interference, and use moving average filtering to process smooth signal fluctuations; Outlier processing: Identify and remove abnormal data through the standard deviation method or box plot method; Signal correction: Calibrate the accelerometer and gyroscope sensors to eliminate signal offset and drift errors;

[0015] S2.2. Normalize the multi-sensor data to eliminate the dimension difference; Achieve time synchronization of multi-source data through timestamp alignment technology to ensure the timing consistency of subsequent analysis;

[0016] S2.3. Time-domain feature extraction: Calculate the statistics of motion parameters: mean, variance, peak-to-peak value; Extract waveform features: Detect motor faults through the integrity detection of the sine waveform of the joint current;

[0017] S2.4. Frequency-domain and time-frequency domain analysis: Analyze the frequency components of the vibration signal through the Fast Fourier Transform (FFT) to identify bearing wear faults; Use the Short-Time Fourier Transform (STFT) to decompose non-stationary signals and extract energy features;

[0018] S2.5. Dimensionality reduction and key feature selection: Use Principal Component Analysis (PCA) to reduce the feature dimension, and retain the projection direction with the largest variance to distinguish normal and abnormal states; Screen the most discriminative features based on the Fisher Discriminant Ratio (FDR) and Spearman correlation coefficient.

[0019] The method for identifying bearing wear faults in S2.4 is as follows: Convert the time-domain signal to a frequency-domain spectrum through the FFT algorithm, calculate the amplitude spectrum and phase spectrum; Use the fftshift function to adjust the spectrum center for easy observation of low-frequency to high-frequency components; Calculate the Bearing Characteristic Frequency (BCF), and the formula is:

[0020]

[0021] where N is the number of balls, f r is the rotational speed frequency, v is the contact angle, and α is the raceway contact angle; Compare the measured spectrum with the theoretical BCF. If there is a significant peak, it is determined as a potential fault.

[0022] The method for decomposing the non-stationary signal and extracting the energy feature in S2.4 is:

[0023] Short-time Fourier transform requires the selection of a suitable window function w(t), which is defined as:

[0024]

[0025] Where L is the window length;

[0026] The original signal x(t) is divided into multiple short-time windows, each with a length of L and an overlapping number of R. Each window segment is expressed as:

[0027] x k (f) = x(t+kR)·w(t)

[0028] Where k represents the window segment number, and t represents the time index;

[0029] For each short-time window x k (f) Perform Fourier transform to obtain the frequency domain signal X k (f), whose expression is:

[0030]

[0031] Where, f represents frequency;

[0032] For each short-time window, the Fourier transform result X k (f), calculate its amplitude spectrum:

[0033]

[0034] Among them, Re(X k (f)) and IM(X k (f)) are the real and imaginary parts respectively;

[0035] Compute the energy spectrum for each short time window:

[0036] E k (f) = 丨X k (f)丨 2

[0037] The energy spectrum reflects the energy distribution of the signal at different frequencies and is an important basis for subsequent feature extraction.

[0038] The method for selecting the most discriminative features based on Fisher's discriminant ratio (FDR) and Spearman's correlation coefficient in S2.5 is:

[0039] The Spearman correlation coefficient is used to measure the monotonic relationship between two variables, and the formula is as follows:

[0040]

[0041] Among them, d i is the difference after sorting two variables in the sample; the Spearman correlation coefficient is used to screen out the features significantly related to the fault severity;

[0042] The Fisher discriminant ratio is used to measure the ratio of between-class variance to within-class variance, and the formula is as follows:

[0043]

[0044] Among them, μ i is the mean of the j-th class, μ is the overall mean, k is the number of classes, and n is the total number of samples;

[0045] The Spearman correlation coefficient and the Fisher discriminant ratio are combined into a mixed score, and the formula is as follows:

[0046] S = ω1·ρ + ω2FDR

[0047] Among them, ω1 and ω2 are weight parameters; all candidate features are sorted according to the mixed score, and the top m features with the highest scores are selected as the final retained feature set.

[0048] The method for evaluating the health status of the robot in S3 is as follows:

[0049] S3.1. Construct a multi-level health parameter system including application level, whole machine level, joint level, etc.; through a comprehensive evaluation algorithm, the health status is quantified into a health score to reflect the overall health status of the robot;

[0050] S3.2. Display the health status of the robot in the form of a radar chart and a bar chart.

[0051] A robot health index intelligent perception and real-time monitoring system includes a data acquisition module, a data processing module, a health evaluation module, and an intervention module. The data acquisition module is connected to the data processing module, the data processing module is connected to the health evaluation module, and the health evaluation module is connected to the intervention module.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] The present invention can collect and process the data of the robot in real time during the task execution, timely detect and quantify the performance degradation, and provide the possibility for timely taking intervention measures. The present invention uses artificial intelligence algorithms for data processing and health evaluation, without manual monitoring and analysis, reducing the labor cost and improving the work efficiency. The present invention eliminates noise and interference through filtering algorithms and machine learning algorithms, extracts key features, and uses deep learning algorithms for health evaluation, improving the accuracy of the evaluation results. Brief Description of the Drawings

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only exemplary. For those of ordinary skill in the art, without creative efforts, other implementation drawings can also be obtained based on the provided drawings.

[0055] The structures, proportions, sizes, etc. illustrated in this specification are only used to cooperate with the content disclosed in the specification for those who are familiar with this technology to understand and read, and are not used to limit the limiting conditions for the implementation of the present invention. Therefore, they do not have substantial technical significance. Any modification of the structure, change in the proportional relationship, or adjustment of the size, without affecting the effects that the present invention can produce and the purposes that can be achieved, should still fall within the scope covered by the technical content disclosed in the present invention.

[0056] Figure 1 is the block diagram of the modules of the present invention;

[0057] Figure 2 is the schematic diagram of the gazebo arena for the experiment of the present invention. Detailed Embodiments

[0058] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. These descriptions are only for further explaining the features and advantages of the present invention, rather than limiting the claims of the present invention; based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope protected by the present application.

[0059] The following will further describe in detail the specific implementation manners of the present invention in combination with the drawings and embodiments. The following embodiments are used to illustrate the present invention, but are not used to limit the scope of the present invention.

[0060] In the description of the present application, it should be noted that unless otherwise clearly defined and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, and it can be the internal communication of two components. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances.

[0061] A method for intelligent perception and real-time monitoring of robot health indicators, such asFigure 1 As shown in the figure, it includes the following steps:

[0062] Step 1: Collect various operation data of the robot in real time during the task execution, including the robot's position information, speed information, acceleration information, power information, sensor data, and the output results of the artificial intelligence system.

[0063] Step 1.1: Install multiple sensors on the robot, including inertial sensors, cameras, lidars, and pressure sensors, for collecting key parameters of the robot's position, speed, acceleration, temperature, and power.

[0064] Step 1.2: Transmit the collected data to the data processing center or cloud server through wireless network or wired connection. For wireless transmission, Wi-Fi, Bluetooth, or a dedicated radio frequency module is used, and for wired transmission, industrial Ethernet is used.

[0065] Step 2: Preprocess and extract features from the collected data. Use filtering algorithms to preprocess the original data to eliminate noise and interference, and extract key features that can reflect the performance degradation of the robot.

[0066] Step 2.1: Filtering algorithm selection: Use Kalman filtering to process non - linear system noise, use low - pass filtering to eliminate high - frequency interference, and use moving average filtering to process smooth signal fluctuations. Outlier processing: Identify and remove abnormal data through the standard deviation method or box - plot method. Signal correction: Calibrate the accelerometer and gyroscope sensors to eliminate signal offset and drift errors.

[0067] Step 2.2: Normalize the multi - sensor data to eliminate the dimension difference. Achieve time synchronization of multi - source data through timestamp alignment technology to ensure the temporal consistency of subsequent analysis.

[0068] Step 2.3: Time - domain feature extraction: Calculate the statistics of motion parameters: mean, variance, peak - to - peak value. Extract waveform features: Detect motor faults through the integrity detection of the sine waveform of the joint current.

[0069] Step 2.4: Frequency - domain and time - frequency domain analysis: Analyze the frequency components of the vibration signal through the Fast Fourier Transform (FFT) to identify bearing wear faults. Use the Short - Time Fourier Transform (STFT) to decompose non - stationary signals and extract energy features.

[0070] Convert the time - domain signal to a frequency - domain spectrum through the FFT algorithm, calculate the amplitude spectrum and phase spectrum. Use the fftshift function to adjust the spectrum center for easy observation of low - frequency to high - frequency components. Calculate the Bearing Characteristic Frequency (BCF), and the formula is:

[0071]

[0072] Where N is the number of balls, f r is the rotational frequency, v is the contact angle, and α is the raceway contact angle. Compare the measured spectrum with the theoretical BCF. If there is a significant peak, it is determined to be a potential fault.

[0073] Short-time Fourier transform requires the selection of a suitable window function w(t), which is defined as:

[0074]

[0075] Where L is the window length.

[0076] The original signal x(t) is divided into multiple short-time windows, each with a length of L and an overlapping number of R. Each window segment is expressed as:

[0077] x k (f) = x(t+kR)·w(t)

[0078] Wherein, k represents the window segment number, and t represents the time index.

[0079] For each short-time window x k (f) Perform Fourier transform to obtain the frequency domain signal X k (f), whose expression is:

[0080]

[0081] Here, f represents frequency.

[0082] For each short-time window, the Fourier transform result X k (f), calculate its amplitude spectrum:

[0083]

[0084] Among them, Re(X k (f)) and IM(X k (f)) are the real and imaginary parts respectively.

[0085] Compute the energy spectrum for each short time window:

[0086] E k (f) = 丨X k (f)丨 2

[0087] The energy spectrum reflects the energy distribution of the signal at different frequencies and is an important basis for subsequent feature extraction.

[0088] Step 2.5, Dimensionality Reduction and Key Feature Selection: Use Principal Component Analysis (PCA) to reduce the feature dimension, and retain the projection direction with the largest variance to distinguish normal and abnormal states. Screen the most discriminative features based on Fisher Discriminant Ratio (FDR) and Spearman correlation coefficient.

[0089] The Spearman correlation coefficient is used to measure the monotonic relationship between two variables, and the formula is as follows:

[0090]

[0091] where d i is the difference between the rankings of the two variables in the sample. Apply the Spearman correlation coefficient to screen out the features that are significantly correlated with the fault severity.

[0092] The Fisher discriminant ratio is used to measure the ratio of between-class variance to within-class variance, and the formula is as follows:

[0093]

[0094] where μ i is the mean of the j-th class, μ is the overall mean, k is the number of classes, and n is the total number of samples.

[0095] Combine the Spearman correlation coefficient and the Fisher discriminant ratio into a mixed score, and the formula is as follows:

[0096] S = ω1·ρ + ω2FDR

[0097] where ω1 and ω2 are weight parameters. Sort all candidate features according to the mixed score, and select the top m features with the highest scores as the final retained feature set.

[0098] Step 3, Based on the artificial intelligence algorithm and the robot health assessment index, evaluate the health status of the robot according to the processed data, and generate the robot health index.

[0099] Step 3.1, Construct a multi-level health parameter system including application level, whole machine level, joint level, etc. Through the comprehensive evaluation algorithm, quantify the health status into a health score to reflect the overall health status of the robot.

[0100] Step 3.2, Display the health status of the robot in the form of a radar chart and a bar chart.

[0101] Step 4, When it is detected that the robot performance drops by more than the preset threshold, automatically or prompt the user to take intervention measures, including adjusting the robot parameters, replacing components or performing remote control.

[0102] An intelligent perception and real-time monitoring system for robot health indicators, including a data acquisition module, a data processing module, a health assessment module, and an intervention module. The data acquisition module is connected to the data processing module, the data processing module is connected to the health assessment module, and the health assessment module is connected to the intervention module.

[0103] Embodiment

[0104]

[0105]

[0106] In two experiments, the robot used the ROS navigation stack with a dynamic window local planner and a global planner using the Djikstra algorithm. Rotational recovery was disabled to minimize confounding factors affecting the experiment. The robot did not receive any prior information about the map, terrain, boundary conditions, or any performance degradation factors. If the robot aborted its navigation plan due to terrain adversity, laser scanner noise, or path planning timeout, the navigation goal was reset. However, if the robot became stuck, unable to find a path to the goal, or aborted navigation for 30 seconds despite goal reset, the experiment trial was terminated.

[0107] Three common field-repairable performance degradation factors were observed in the harsh environment: high-friction (HF) terrain, uneven terrain, and laser noise. Laser scanner noise reduces the robot's ability to perceive the environment, thus introducing localization and navigation errors. Crossing uneven terrain causes instability. Height variations may cause the laser scanner to detect the ground as an obstacle, thus reducing navigation ability. Finally, the robot faces difficulties in turning and smooth movement on HF terrain. On such ground, the robot may slip, skid, or even stop. In this case, the values of d˙g, x,˙, and δ˙loc are all affected. Different intensities of these performance degradation factors were combined to form multiple experimental conditions.

[0108] In Experiment 1 and Experiment 2, 15 and 14 trials were conducted for each degradation level, respectively. In addition, in this embodiment, the health value of the robot was extracted, measured during operation, and the average health value of the robot for each experimental trial was calculated. Performance degradation increases the time required for the robot to complete the task. Therefore, in this embodiment, the time to complete the navigation task (Tcomp) was used as an objective post-hoc measure of performance degradation. By increasing the degree of performance degradation in the task, it was hypothesized that: 1) the value of Tcomp increases, 2) the average health value of the robot during the experimental run decreases, and 3) the average health value of the robot is negatively correlated with Tcomp (i.e., as the health value of the robot decreases, the value of Tcomp increases).

[0109] A. Experiment 1

[0110] As Figure 2 shown, Experiment I used Gazebo, a high-fidelity robot simulation with a realistic physics engine. The robot was equipped with wheel encoders for odometry, an LMS-111 lidar scanner, and a UM6 IMU sensor. Seven seconds after the start of each trial, different levels of random additive Gaussian white noise were introduced into the lidar scanner, and the Box-Muller transform was used to degrade the robot's localization ability. The lidar noise was turned off after 7 seconds. The duration of the noise was heuristically chosen so that sufficient performance degradation would occur during operation without causing robot failure. To control the magnitude of the noise, the standard deviation of the Gaussian kernel was multiplied by the noise scale. For Experiment I, four different arenas with uneven terrains were also designed in this embodiment. The total areas of these arenas were covered with "FRC 2016 Rough Terrain" pavilion blocks at 0%, 10%, 20%, and 40% respectively.

[0111] Results of Experiment I: The performance degradation increased the average time required for the robot to complete the task and decreased the average health value during operation. However, for different levels of performance degradation, the ranges of values for Tcomp and the average health value of the robot were different. The performance degradation caused by the combination of lidar noise and uneven terrain was greater than that of any single factor alone. However, the experiments provided evidence that the effects of multiple performance degradation factors on the robot may not always be additive in nature, and quantifying the effects of performance degradation factors into different levels is not trivial. However, their intensities can be measured using relative values. Among the different performance degradation factors, the degradation caused by lidar noise was the lowest because it caused the smallest increase in Tcomp compared to the baseline performance. The presence of noise and uneven terrain resulted in the lowest range of average health values (-1.22 to -

[0112] 1.484). Finally, the Spearman rank correlation test showed a significant negative correlation between Tcomp and the average health status of the robot (p < 0.001, ρ = -0.93).

[0113] B. Experiment II

[0114] Experiment 2 was conducted using a real robot. The robot used in this experiment did not have an IMU, so the value of \(\dot{a}_z\) was not calculated. The health of the robot was calculated using \(V = \{\dot{d}_g, \dot{\delta}_{loc,x}, \dot{\sigma}_{noise}^2\}\) from Equation 5. To avoid the risk of hardware damage caused by the robot tipping over, uneven terrain was not used in the real robot experiment. Instead, HF terrain and obstacles were used to reduce the robot's ability to turn and move smoothly. The HF terrain consisted of square floor tiles covered with high-friction rubber pads. These tiles were not fixed to the floor because the sliding of the tiles when the robot turned would cause positioning errors, which increased the performance degradation.

[0115] The range of \(T_{comp}\) values increased as the degree of performance degradation increased. The interquartile range of the average health observed at level 3 was lower than that at level 2, but the range of values was similar. The robot was unable to complete the task in 2 trials at level 2 and 3 trials at level 3. In these trials, the robot temporarily experienced positioning errors on the HF terrain and then could not find a collision-free path even after resetting the goal. The Spearman rank correlation test showed a significant (p < 0.001), strong negative correlation (\(\rho=-0.77\)) between the average health of the robot and the \(T_{comp}\) value. Given the same experimental conditions, robot hardware, and navigation algorithm, the performance degradation caused by the robot varied in each trial. The health value trend at level 1 decreased from 20 seconds to 30 seconds. During this time, the robot encountered obstacles and slowed down to create a new navigation plan. Then the health continued to rise until the task was completed, indicating little performance degradation. At level 2, the health value dropped sharply around 50 seconds. This sharp drop was consistent with the robot encountering HF terrain and facing navigation errors. At level 3, the health value trend of the robot was characterized by high fluctuations due to laser noise and the influence of obstacles. The health trends (L to R) under all conditions in Experiment 2: level 1, level 2, level 3.

[0116] The evidence provided shows that as the degree of performance degradation increases, \(T_{comp}\) increases and the average health of the robot decreases. In addition, the health of the robot successfully tracked the performance degradation in real time. Most importantly, the strong negative correlation between \(T_{comp}\) and the average health of the robot indicates that the proposed online metric for measuring performance degradation in this embodiment is as effective as the objective post hoc measurement (i.e., \(T_{comp}\)). The evidence provided shows that the health of the robot can successfully estimate the instantaneous performance degradation of the robot during operation.

[0117] An online robot performance monitoring system can be implemented within this framework. Such a system can autonomously detect low health (i.e., high performance degradation) situations, trigger recovery behaviors, or request operator assistance. Experiment II gives insights into how such a system can be proven useful. In some trials of Experiment II, the robot would get temporarily stuck or fail when it could not find a collision-free path through the arena. Thus, a simple threshold-based control switcher can be used to initiate recovery behaviors. Alternatively, the operator can control the robot, providing it with a collision-free path or operating it remotely. In multi-robot applications, robot health can also be used to prioritize robots that require operator attention based on the severity of performance degradation. Explainable AI agents can assist the operator and autonomous robots in initiating recovery behaviors, and the robot's vital signs and health can be used to validate their decisions. However, elaborating on the design and implementation of such a control switcher is beyond the scope of this embodiment. Although the health metrics in Experiment I used 5 vital signs, Experiment I did not use a˙z because the robot did not have an IMU. The framework of this embodiment is still able to capture the overall impact of robot performance degradation, indicating its generality, scalability, and robustness for adding or removing at least one crucial metric. Any new metric that conforms to the robot's vital signs can be easily added to the health metrics using the P(suffering|v) function, which relates changes in the vital signs to the probability of robot failure. This function can be derived from Monte Carlo studies, reliability studies, or expert knowledge. For example, expert knowledge can be used to identify incorrect sensor readings or unusual robot behaviors in a specific environment. Alternatively, methods such as principal component analysis or machine learning can be used to find a set of vital signs. These methods can mine large amounts of data from robot operations to find the parameters that best represent its performance degradation. Similar techniques can also be used to design the P(pain) function or health metric. However, encoding expert knowledge in this system can be intuitively interpreted compared to black-box artificial intelligence. Some differences in results are due to the randomness introduced by performance degradation factors in the experiment. During low performance degradation or normal operation, minor differences in LIDAR map representation, odometry, or update frequency do not affect the robot's navigation. However, factors such as laser noise can distort the robot's perception of the environment. In some trials, this can lead to an increased likelihood of collisions and failures; in other trials, the robot follows a suboptimal path towards the goal. In Experiment II, minor differences in the robot's path caused it to climb at different angles on high-friction terrain. In some of these cases, the robot got stuck / experienced failure, while in other cases, it was able to complete the task. Due to this randomness, an increase in the level of performance degradation in the experiment does not always result in a corresponding increase in Tcomp.

[0118] The above only elaborates in detail on the preferred embodiments of the present invention. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those of ordinary skill in the art, various changes can be made without departing from the gist of the present invention, and all such changes should be included within the scope of protection of the present invention.

Claims

1. An intelligent perception and real-time monitoring method for robot health indicators, characterized in that, Including the following steps: S1. Collect various operation data of the robot in real time during task execution, including the robot's position information, speed information, acceleration information, power information, sensor data, and the output results of the artificial intelligence system; S2. Preprocess and extract features from the collected data. Use a filtering algorithm to preprocess the original data to eliminate noise and interference, and extract key features that can reflect the performance degradation of the robot; S3. Based on artificial intelligence algorithms and robot health assessment metrics, evaluate the health status of the robot according to the processed data and generate robot health metrics; S4. When it is detected that the robot's performance degradation exceeds a preset threshold, automatically or prompt the user to take intervention measures, including adjusting robot parameters, replacing components, or performing remote control.

2. The intelligent perception and real-time monitoring method for the health indicators of a robot according to claim 1, characterized in that, The method for collecting various operation data of the robot in real time in S1 is as follows: S1.

1. Install a variety of sensors on the robot, including inertial sensors, cameras, lidar, and pressure sensors, for collecting key parameters of the robot's position, speed, acceleration, temperature, and power; S1.

2. Transmit the collected data to a data processing center or cloud server through wireless or wired means. The wireless transmission uses Wi-Fi, Bluetooth, or a dedicated radio frequency module, and the wired transmission uses industrial Ethernet.

3. The intelligent perception and real-time monitoring method for robot health indicators according to claim 1, characterized in that The method for preprocessing the original data in S2 is as follows: S2.

1. Filtering algorithm selection: Use Kalman filtering to process non-linear system noise, use low-pass filtering to eliminate high-frequency interference, and use moving average filtering to process smooth signal fluctuations; Outlier processing: Identify and remove abnormal data through the standard deviation method or box plot method; Signal correction: Calibrate the accelerometer and gyroscope sensors to eliminate signal offset and drift errors; S2.

2. Normalize the multi-sensor data to eliminate dimensional differences; Achieve time synchronization of multi-source data through timestamp alignment technology to ensure the temporal consistency of subsequent analysis; S2.

3. Time-domain feature extraction: Calculate the statistics of motion parameters: mean, variance, peak-to-peak value; Extract Waveform features: Detect motor faults through the integrity detection of the sine waveform of joint current; S2.

4. Frequency-domain and time-frequency domain analysis: Analyze the frequency components of vibration signals through the Fast Fourier Transform (FFT) to identify bearing wear faults; Use the Short-Time Fourier Transform (STFT) to decompose non-stationary signals and extract energy features; S2.

5. Dimensionality reduction and key feature selection: Use Principal Component Analysis (PCA) to reduce the feature dimension, and retain the projection direction with the largest variance to distinguish normal and abnormal states; Screen the most discriminative features based on the Fisher Discriminant Ratio (FDR) and Spearman correlation coefficient.

4. The intelligent perception and real-time monitoring method for the health indicators of a robot according to claim 3, characterized in that, The method for identifying bearing wear faults in S2.4 is as follows: Convert the time-domain signal to a frequency-domain spectrum through the FFT algorithm, calculate the amplitude spectrum and phase spectrum; Use the fftshift function to adjust the spectrum center for easy observation of low-frequency to high-frequency components; Calculate the Bearing Characteristic Frequency (BCF), and the formula is: where N is the number of balls, f r is the rotational speed frequency, v is the contact angle, and α is the raceway contact angle; comparing the measured spectrum with the theoretical BCF, if there is a significant peak, it is determined as a potential fault.

5. The intelligent perception and real-time monitoring method for the health indicators of a robot according to claim 3, characterized in that, The method for decomposing non-stationary signals and extracting energy features in S2.4 is as follows: The short-time Fourier transform requires the selection of a suitable window function w(t), and the definition of the window function is as follows: where L is the window length; The original signal x(t) is divided into multiple short-time windows, each with a length of L and an overlap of R points. Each window segment is expressed as: x k f(t) = x(t + kR)·w(t) where k represents the window segment number and t represents the time index; For each short-time window \(x\) k (f), perform Fourier transform to obtain the frequency-domain signal \(X\) k (f), and its expression is: where f represents the frequency; For the Fourier transform result X k (f) of each short time window, calculate its magnitude spectrum: where Re(X k (f)) and IM(X k (f)) are the real part and the imaginary part, respectively; Calculate the energy spectrum of each short-time window: E k f) = |X k (f)| 2 The energy spectrum reflects the energy distribution of the signal at different frequencies and is an important basis for subsequent feature extraction.

6. The intelligent perception and real-time monitoring method for the health indicators of a robot according to claim 3, characterized in that, The method for screening the most discriminative features based on the Fisher Discriminant Ratio (FDR) and Spearman correlation coefficient in S2.5 is as follows: The Spearman correlation coefficient is used to measure the monotonic relationship between two variables, and the formula is as follows: where d i is the difference between the rankings of two variables in the sample; the Spearman correlation coefficient is applied to screen out the features that are significantly correlated with the severity of the fault; The Fisher discriminant ratio is used to measure the ratio of the between-class variance to the within-class variance, and the formula is as follows: where μ i is the mean of the j-th class, μ is the overall mean, k is the number of classes, and n is the total number of samples; Combine the Spearman correlation coefficient and the Fisher discriminant ratio into a mixed score, and the formula is as follows: S = ω1·ρ + ω2FDR where ω1 and ω2 are weight parameters; sort all candidate features according to the mixed score, and select the top m features with the highest scores as the final retained feature set.

7. A method for intelligent perception and real-time monitoring of a robot's health indicators according to claim 1, characterized in that, The method for evaluating the health status of the robot in S3 is as follows: S3.

1. Construct a multi-level health parameter system including application level, whole machine level, joint level, etc.; through a comprehensive evaluation algorithm, quantify the health status into a health score to reflect the overall health status of the robot; S3.

2. Display the health status of the robot in the form of a radar chart and a bar chart.

8. An intelligent perception and real-time monitoring system for robot health indicators, characterized in that, It includes a data acquisition module, a data processing module, a health evaluation module, and an intervention module. The data acquisition module is connected to the data processing module, the data processing module is connected to the health evaluation module, and the health evaluation module is connected to the intervention module.

Images