A waterwheel type loading and unloading robot fault diagnosis method and system

Abstract

The application discloses a waterwheel type loading and unloading robot fault diagnosis method and system, the method comprises the following steps: acquiring the characteristic data of the waterwheel type loading and unloading robot in each normal state, and preprocessing the characteristic data to form a multi-dimensional characteristic state set; determining the threshold range of the waterwheel type loading and unloading robot in each normal state according to the multi-dimensional characteristic state set; comparing the characteristic data of the waterwheel type loading and unloading robot to be diagnosed with the threshold range after acquiring the characteristic data, and determining the fault state of the waterwheel type loading and unloading robot to be diagnosed according to the comparison result; after analyzing the data of the waterwheel type loading and unloading robot in the fault state, the multi-dimensional state deviation degree algorithm based on the weighted Euclidean distance is used for comprehensive judgment, a fault positioning report and a processing strategy are generated, and real-time display is performed through a visual interface. The application improves the diagnosis accuracy and prediction ability, and provides reliable guarantee for the safe and stable operation of the waterwheel type loading and unloading robot.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology for waterwheel-type loading and unloading robots, and specifically to a fault diagnosis method and system for waterwheel-type loading and unloading robots. Background Technology

[0002] As a core intelligent equipment in loading and unloading operations, the continuous and stable operation of water-wheel-type loading and unloading robots is crucial to ensuring operational efficiency and reducing operating costs. Fault diagnosis technology is a core component of the robot equipment operation and maintenance system, directly determining the efficiency of equipment fault diagnosis, the accuracy of fault warning, and the formulation of subsequent maintenance strategies.

[0003] Currently, the main fault diagnosis methods for water-wheel loading and unloading robots in the industry include expert system diagnosis, neural network diagnosis, Bayesian network diagnosis, and hybrid model diagnosis. Among these, expert system diagnosis often faces difficulties in knowledge acquisition, incomplete knowledge representation, and poor adaptability to new faults. Neural network diagnosis requires a large amount of data for training, and the network structure is often complex, resulting in long training times. It is also sensitive to noisy data, and small errors or outliers can cause the diagnostic results to deviate from the correct direction. Bayesian network diagnosis requires manual parameter adjustment, making it difficult to find optimal parameters, and the accuracy of Bayesian networks largely depends on the quality and completeness of prior knowledge. Hybrid model diagnosis requires fusing the outputs of different models, necessitating the discovery of a suitable fusion strategy, and is prone to overfitting.

[0004] At the same time, although the loading and unloading industry has established a general technical architecture of perception-transmission-processing-application, integrating technologies such as IoT perception, edge computing and 5G communication, AI intelligent diagnosis and digital twins, the existing technical system is mostly designed for general loading and unloading equipment. It has not been customized and adapted to the structural characteristics and operating rules of water-wheel loading and unloading robots. It cannot accurately extract the feature correlation of core operating parameters, making it difficult to accurately identify and warn of potential early equipment failures. Moreover, it still cannot get rid of the dependence on post-event alarms and manual inspections. The passive maintenance mode still dominates, which can easily cause equipment downtime, operation interruption, and even secondary damage to equipment due to untimely fault detection. Summary of the Invention

[0005] In order to solve the above-mentioned technical problems, this application proposes the following technical solution: In a first aspect, embodiments of this application provide a fault diagnosis method for a waterwheel-type loading and unloading robot, including: The feature data of the waterwheel loading and unloading robot under various normal states are acquired, and the feature data is preprocessed to form a multi-dimensional feature state set. The threshold range of the waterwheel loading and unloading robot under each normal state is determined based on the multidimensional feature state set. After acquiring the feature data of the water-wheel loading and unloading robot to be diagnosed, it is compared with the threshold range, and the fault status of the water-wheel loading and unloading robot to be diagnosed is determined based on the comparison results. After analyzing the data of the waterwheel loading and unloading robot that is determined to be in a faulty state, a multi-dimensional state deviation algorithm based on weighted Euclidean distance is used to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

[0006] In one possible implementation, acquiring feature data of the waterwheel-type loading and unloading robot under various normal states, and preprocessing the feature data to form a multi-dimensional feature state set, includes: Vibration sensors are used to collect the raw three-axis acceleration time series signals of the water-wheel loading and unloading robot at a high frequency sampling rate. After filtering out noise through a digital filter, the noise-reduced signal is standardized. The time-domain features are calculated from the standardized time-series original signal, and the signal is transformed to the frequency domain by fast Fourier transform and the frequency domain features are calculated. The temporal and frequency domain features are combined to form a vibration feature vector; Torque signals are collected during the operation of the water-wheel loading and unloading robot by a torque sensor. The complete working cycle is divided by key point detection to obtain a periodic torque sequence. The dynamic time warping algorithm is used to align multiple cycles in time and calculate the statistical and shape features in each aligned cycle. The statistical and shape features are then combined to form a torque feature vector. Temperature data of key parts of the water-wheel loading and unloading robot are collected by temperature sensors, and the trend signal is obtained by smoothing with a moving average filter. The standard deviation within the sliding window is then calculated. The standard deviation is used to determine whether the water-wheel loading and unloading robot has entered the thermal steady-state working stage; The average value within the steady-state window is calculated as the core temperature variable, and the temperature rise rate from startup to steady state is calculated. The core temperature variable and the temperature rise rate are combined to form a temperature feature vector. The strain signals of key parts of the waterwheel loading and unloading robot are collected by strain sensors, obvious abnormal jump points are eliminated, and the collected voltage signals are converted into micro-strain physical quantities according to the bridge formula and sensor sensitivity coefficient. The average strain and maximum strain during the monitoring period are calculated. The amplitude-mean spectrum of strain fluctuations is statistically analyzed using the rainflow counting method to obtain the fatigue damage degree. The average strain, maximum strain, and fatigue damage are combined to form a strain feature vector; The vibration feature vector, torque feature vector, temperature feature vector, and strain feature vector are aligned and fused within a unified time window to form a multidimensional feature state set.

[0007] In one possible implementation, the threshold range of the waterwheel loading and unloading robot under various normal states is determined based on the multidimensional feature state set, including: The mean and standard deviation of each feature parameter in the multidimensional feature state set are calculated using statistical analysis methods. The threshold range is set based on the calculated mean and standard deviation, and the threshold range is dynamically adjusted according to the operating conditions of the waterwheel loading and unloading robot.

[0008] In one possible implementation, the step of acquiring the feature data of the waterwheel loading and unloading robot to be diagnosed and comparing it with the threshold range, and determining the fault status of the waterwheel loading and unloading robot to be diagnosed based on the comparison result, includes: Obtain the current operating status of the waterwheel loading and unloading robot, and select the corresponding threshold range based on the current status; After determining the current state, the vibration data, temperature data, torque data, and strain data of the water-wheel loading and unloading robot to be diagnosed are obtained respectively. The vibration data, temperature data, torque data, and strain data are preprocessed to obtain the feature vector of the current state; The feature vector of the current state is compared with the corresponding threshold range to determine whether each feature parameter is within the threshold range. If all feature parameters are within the threshold range, the waterwheel loading and unloading robot is determined to be in normal condition. If any feature parameter exceeds the threshold range, the waterwheel loading and unloading robot is determined to be in a fault state. If multiple characteristic parameters exceed the threshold range simultaneously, the waterwheel loading and unloading robot is determined to be in a serious malfunction state.

[0009] In one possible implementation, the analysis of data from the waterwheel-type loading and unloading robot determined to be in a fault state includes: Real-time acquisition of vibration, temperature, torque, and strain data under fault conditions; The vibration data is analyzed using a vibration analysis algorithm to diagnose the health status of the bearing. The temperature data is analyzed using a temperature analysis algorithm to identify abnormal operating conditions and determine the abnormal measurement points and causes. The torque data is analyzed using a torque analysis algorithm to diagnose the condition of the spindle; The strain data is analyzed using a strain analysis algorithm to assess the deformation of key components.

[0010] In one possible implementation, the step of analyzing the vibration data using a vibration analysis algorithm to diagnose the health status of the bearing includes: Based on the vibration data, sampling rate, and bearing parameters, the cage frequency, inner ring failure frequency, outer ring failure frequency, rolling element failure frequency, and root mean square value are calculated respectively, using the following formulas: in, For frequency conversion, The shaft speed is To maintain the rack frequency, The diameter of the rolling element, The diameter of the pitch circle. Contact angle, For the inner ring fault frequency, For the number of rolling elements, For the outer ring fault frequency, For the rolling element failure frequency, The number of sampling points. For the first One vibration acceleration data point, It is the root mean square value; The frequency matching range is determined based on the calculated theoretical bearing fault characteristic frequencies. The calculation formula is as follows: in, To match the frequency range, This represents the theoretical characteristic frequency of bearing failure. The vibration fault is determined based on the matched frequency range combined with the detected peak frequency, peak amplitude, and amplitude detection threshold. The calculation formula is as follows: in, The detected peak frequency, The detected peak amplitude, The amplitude detection threshold; The vibration RMS penalty is determined based on the root mean square value, and the penalty fault is determined based on the detected fault frequency. The vibration health score is calculated based on the vibration RMS penalty and the penalty fault, using the following formula: in, The peak amplitude corresponding to the detected fault frequency; The remaining life of the vibration bearing is estimated based on the vibration health score.

[0011] In one possible implementation, the step of analyzing the torque data using a torque analysis algorithm to diagnose the spindle's condition includes: The mean, standard deviation, and deviation of the torque are calculated based on the torque data, sampling rate, and rated torque, using the following formulas: in, The average torque. The number of sampling points. For the first One torque data point, For the standard deviation of torque, This is the rated torque; Volatility is calculated based on the mean and standard deviation, and the trend of torque variation is analyzed. The calculation formulas are as follows: in, The regression slope, For the first One torque value, For time indexing; When the calculated volatility exceeds a preset threshold, the periodicity of the torque signal fluctuation is analyzed. The health score of torque is calculated based on the aforementioned deviation, volatility, regression slope, and periodicity, using the following formula: The condition of the spindle is determined based on the torque health score.

[0012] In one possible implementation, the step of analyzing the strain data using a strain analysis algorithm to evaluate the deformation of key components includes: The safety factor and strain exceedance rate are calculated based on the strain data, strain history data, and strain limits, respectively, using the following formulas: in, For strain limits, To prepare for the current situation; Simultaneously, after analyzing the strain variation trend, the correlation between strains at different measuring points was analyzed, and the calculation formulas are as follows: in, The slope of the strain regression. For the first The time of the measurement For the first The strain value measured in this test. To measure the number of times, The Pearson correlation coefficient is used. For the measurement point A, the first This measurement value, For the measurement point B, the first The measurement value; The strain health score is calculated based on the over-limit rate, trend judgment threshold, strain standard deviation, and strain standard deviation threshold. The calculation formula is as follows: in, Points are awarded for exceeding the limit. This is the penalty coefficient for exceeding the limit. Punishment for adapting to trends The threshold for trend determination. To compensate for fluctuations, For strain standard deviation, The threshold for triggering fluctuations, The standard deviation threshold for strain. For the mean strain, For the number of measurements; The corresponding strain state is classified according to the strain health score.

[0013] In one possible implementation, a multi-dimensional state deviation algorithm based on weighted Euclidean distance is used for comprehensive judgment, generating a fault location report and handling strategy, which is then displayed in real time through a visual interface, including: Define the current state vector and the normal baseline state vector; The overall deviation is calculated based on the current state vector, the normal baseline state vector, and the standard deviation of each feature under normal conditions. The calculation formula is as follows: Where n is the total dimension of the features. Let this be the current state vector. This is the normal baseline state vector. The standard deviation of each feature under normal conditions. For the weights of each dimension, This refers to the overall deviation. Based on the comprehensive deviation, a final fault location report and handling strategy are generated and displayed in real time through a visual interface.

[0014] Secondly, embodiments of this application provide a fault diagnosis system for a waterwheel-type loading and unloading robot, including: The acquisition module is used to acquire feature data of the waterwheel loading and unloading robot under various normal states, and to preprocess the feature data to form a multi-dimensional feature state set. The determination module is used to determine the threshold range of the waterwheel loading and unloading robot under each normal state based on the multidimensional feature state set; The comparison and judgment module is used to acquire the feature data of the water-wheel loading and unloading robot to be diagnosed and compare it with the threshold range, and judge the fault status of the water-wheel loading and unloading robot to be diagnosed based on the comparison result. The fault diagnosis module analyzes the data of the waterwheel loading and unloading robot that is determined to be in a faulty state, and then uses a multi-dimensional state deviation algorithm based on weighted Euclidean distance to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

[0015] Compared with the prior art, the beneficial effects of this application are as follows: This application addresses the core operating parameters of water-wheel loading and unloading robots by developing specific feature extraction and preprocessing schemes for four-dimensional signals: vibration, torque, temperature, and strain. Through time-frequency domain analysis, it captures early damage characteristics of transmission components such as bearings and gears from vibration signals. Periodic segmentation and shape analysis reflect real-time load anomalies and transmission efficiency declines from torque signals. Steady-state and dynamic feature extraction monitors potential faults such as motor overheating and insufficient cooling efficiency from temperature signals. Stress statistics and fatigue analysis provide early warnings of structural problems such as overload and fatigue cracks from strain signals. Based on fused features, it calculates the deviation between the current equipment state and the baseline health state in real time, enabling the identification of potential abnormal trends before significant equipment failures occur. This allows for accurate early warnings of faults, facilitating maintenance personnel to develop proactive maintenance strategies and eliminate faults at their inception. This model fundamentally changes the passive approach of traditional reactive maintenance, effectively reducing downtime and operational losses caused by sudden equipment failures. It also avoids secondary damage caused by escalating faults, significantly reducing equipment maintenance costs and overall lifecycle maintenance costs, achieving a strategic upgrade to data-driven predictive maintenance and intelligent prevention. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a fault diagnosis method for a waterwheel-type loading and unloading robot provided in an embodiment of this application; Figure 2 A flowchart of the vibration analysis algorithm provided in the embodiments of this application; Figure 3 A flowchart of the temperature analysis algorithm provided in this application embodiment; Figure 4 A flowchart of the torque analysis algorithm provided in the embodiments of this application; Figure 5 A flowchart of the strain analysis algorithm provided in the embodiments of this application; Figure 6 A schematic diagram of a fault diagnosis system for a waterwheel-type loading and unloading robot provided in an embodiment of this application; Figure 7 This is a diagram of a LangGraph-based closed-loop architecture provided for an embodiment of this application. Detailed Implementation

[0017] The present solution will now be described in conjunction with the accompanying drawings and specific embodiments.

[0018] Figure 1 A flowchart illustrating a fault diagnosis method for a waterwheel-type loading and unloading robot provided in this application embodiment is shown below. Figure 1 A fault diagnosis method for a waterwheel-type loading and unloading robot in this embodiment includes: S101: Acquire feature data of the waterwheel loading and unloading robot under various normal states, and preprocess the feature data to form a multi-dimensional feature state set.

[0019] In this embodiment, a vibration sensor is used to collect the original three-axis acceleration time series signal of the waterwheel loading and unloading robot at a high-frequency sampling rate. A digital filter is used to remove high-frequency noise and low-frequency drift, retaining the key frequency band from 1Hz to five times the machine's rotation frequency. The denoised signal is then Z-score normalized to eliminate the influence of different dimensions and sensor sensitivity differences. Time-domain features, including root mean square value, peak value, kurtosis, and skewness, are calculated from the normalized time series. The signal is then converted to the frequency domain using a fast Fourier transform, and frequency-domain features, including dominant frequency energy, spectral centroid, and spectral entropy, are calculated. These N features are combined into a feature vector representing the vibration mode of the current state.

[0020] Torque signals are collected during the operation of the water-wheel loading and unloading robot using torque sensors. The complete work cycle is segmented using key point detection sensors such as torque peak values ​​and angle sensor signals to obtain a periodic torque sequence. A dynamic time warping algorithm is then used to align multiple cycles, eliminating minor time fluctuations. Statistical characteristics within each aligned cycle are calculated, including peak value, mean, and standard deviation. Shape characteristics are also calculated, including the area under the torque-time curve and the torque value at a specific phase point. These features are combined into a torque feature vector representing load size, fluctuations, and work done.

[0021] Temperature data from key components of the water-wheel loading and unloading robot is collected using temperature sensors. A trend signal is obtained by smoothing the data using a moving average or low-pass filter. The standard deviation within the moving window is calculated. When the standard deviation remains below a threshold for a period of time, the robot is considered to have entered a thermal steady-state operating phase. The average value within the steady-state window is calculated as the core temperature variable, reflecting the heating status of the motor and bearings. The temperature rise rate from startup to steady state is also calculated, indirectly reflecting cooling efficiency or load changes. The core temperature variable and the temperature rise rate are combined to form a temperature feature vector.

[0022] Strain signals from key structural components of the waterwheel loading and unloading robot are collected using strain sensors. Obvious anomalous jumps are eliminated. Based on the bridge formula and sensor sensitivity coefficient, the collected voltage signals are converted into micro-strain physical quantities. The average strain and maximum strain during the monitoring period are calculated to reflect the structural stress level under static load. The amplitude-mean spectrum of strain fluctuations is statistically analyzed using the rainflow counting method or range pair method to obtain the fatigue damage degree. The average strain, maximum strain, and fatigue damage degree are combined to form a strain characteristic vector.

[0023] The vibration feature vector, torque feature vector, temperature feature vector and strain feature vector are aligned and fused within a unified time window to form a multidimensional feature state set.

[0024] S102, Determine the threshold range of the waterwheel loading and unloading robot under each normal state based on the multi-dimensional feature state set.

[0025] In this embodiment, statistical analysis methods are used to calculate the mean and standard deviation of each feature parameter in the multidimensional feature state set, and a threshold range is set. For key feature parameters, a more stringent threshold range is adopted. The threshold range is dynamically adjusted based on the operating conditions of the waterwheel-type loading and unloading robot, such as load, ambient temperature, and operating speed. Machine learning algorithms, such as support vector machines and random forests, are used to establish a mapping relationship between operating conditions and threshold ranges, thereby achieving adaptive adjustment of the threshold.

[0026] S103: After acquiring the feature data of the water-wheel loading and unloading robot to be diagnosed, compare it with the threshold range, and determine the fault status of the water-wheel loading and unloading robot to be diagnosed based on the comparison results.

[0027] In this embodiment, the current operating status of the waterwheel loading and unloading robot is obtained, and a corresponding threshold range is selected based on the current status. After determining the current status, vibration data, temperature data, torque data, and strain data of the waterwheel loading and unloading robot to be diagnosed are obtained respectively. The vibration data, temperature data, torque data, and strain data are preprocessed to obtain the feature vector of the current status. The feature vector of the current status is compared with the corresponding threshold range to determine whether each feature parameter is within the threshold range. If all feature parameters are within the threshold range, the waterwheel loading and unloading robot is determined to be in a normal state. If any feature parameter exceeds the threshold range, the waterwheel loading and unloading robot is determined to be in a fault state. If multiple feature parameters exceed the threshold range at the same time, the waterwheel loading and unloading robot is determined to be in a serious fault state.

[0028] S104 analyzes the data of the water-wheel loading and unloading robot that is determined to be in a faulty state, and then uses a multi-dimensional state deviation algorithm based on weighted Euclidean distance to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

[0029] See Figure 2In this embodiment, after a fault condition is determined, vibration data, temperature data, torque data, and strain data under the fault condition are acquired in real time. The vibration data is analyzed using a vibration analysis algorithm to diagnose the health status of the bearing. In this embodiment, the vibration analysis algorithm is based on the bearing fault characteristic frequency theory and can identify faults in the bearing's inner ring, outer ring, rolling elements, and cage, and assess the bearing's remaining life. Based on the vibration data, sampling rate, and bearing parameters, the cage frequency, inner ring fault frequency, outer ring fault frequency, rolling element fault frequency, and root mean square value are calculated respectively. The calculation formulas are as follows: in, For frequency conversion, The shaft speed is To maintain the rack frequency, The diameter of the rolling element, The diameter of the pitch circle. Contact angle, For the inner ring fault frequency, For the number of rolling elements, For the outer ring fault frequency, For the rolling element failure frequency, The number of sampling points. For the first One vibration acceleration data point, This is the root mean square value.

[0030] The Fast Fourier Transform (FFT) is used to convert time-domain signals into frequency-domain signals to analyze frequency components. The FFT formula is: in, For the first The complex values ​​of each frequency component For the first Each time-domain sampling point For frequency index, The imaginary unit is used; the formula for calculating frequency resolution is: in, For frequency resolution, The sampling rate.

[0031] The frequency axis is calculated based on the sampling rate, and the calculation formula is as follows: Fault frequency detection is achieved by finding peak values ​​near the theoretical fault frequency. The frequency matching range is determined based on the calculated theoretical bearing fault characteristic frequency, and the calculation formula is as follows: in, To match the frequency range, This represents the theoretical characteristic frequency of bearing failure. Vibration faults are determined based on the matching frequency range, the detected peak frequency, peak amplitude, and amplitude detection threshold. The calculation formula is as follows: in, The detected peak frequency, The detected peak amplitude, The amplitude detection threshold; The vibration RMS penalty is determined based on the root mean square value, and the penalty fault is determined based on the detected fault frequency. The vibration health score is calculated based on the vibration RMS penalty and the penalty fault, using the following formula: in, The remaining life of the vibratory bearing is estimated based on the peak amplitude corresponding to the detected fault frequency and the vibration health score.

[0032] Temperature data is analyzed using temperature analysis algorithms to identify abnormal operating conditions and determine the abnormal measurement points and their causes. (See also...) Figure 3 In this embodiment, abnormal operating conditions such as poor contact and overload are identified through multi-point temperature monitoring, supporting temperature comparison analysis, trend tracking and change prediction.

[0033] The temperature rise and temperature ratio are calculated based on the current temperature and the reference temperature. The temperature change is calculated based on the most recent and earliest measured temperatures. If the temperature change is greater than 2 degrees Celsius, it is considered an upward trend; if the temperature change is less than -2 degrees Celsius, it is considered a downward trend. The calculation formulas are as follows: in, The current temperature. As the reference temperature, For the temperature rise, The change in temperature The temperature most recently measured. The earliest temperature measured.

[0034] An abnormality is determined when the current temperature exceeds the temperature alarm threshold. The degree of abnormality is determined as follows: Subsequently, a comparative analysis of multiple temperature points was conducted. Temperature health score. The calculation formula is: in: The over-temperature penalty reflects the degree to which the current temperature exceeds the alarm threshold. This is a trend penalty, reflecting the risk of a continued rise or fall in temperature. The deviation penalty reflects the degree of temperature rise deviating from the normal baseline.

[0035] The over-temperature penalty judgment condition is: when the temperature... Temperature exceeds preset alarm threshold If the temperature exceeds the limit, an overheating penalty is triggered; otherwise, the penalty is 0.

[0036] Calculation formula: The criterion for determining the trend penalty is: based on the most recent temperature measurement. Compared with the earliest temperature measurement Change Make a judgment when When, when, it is determined to be a continuous upward trend, when At that time, it was judged to be a stable trend.

[0037] Calculation formula: The criteria for determining deviation penalty are: based on temperature. , in This is the reference temperature under normal conditions. When the temperature rise exceeds the preset allowable temperature rise threshold... For example If the deviation penalty is triggered, the penalty is 0; otherwise, the penalty is 0.

[0038] Calculation formula: Calculate Then, the temperature health status can be classified according to the following conditions: when When the temperature is in good condition, the temperature control is considered normal; when... When the temperature is normal, the patient's health status is considered good, and normal temperature is recommended; when... At that time, the temperature health status is judged as "Caution," indicating a potential risk of overheating; when If the temperature is below 50°C, the temperature health status is considered serious, and the cooling system or load should be checked immediately.

[0039] Torque analysis algorithms are used to analyze torque data and diagnose the condition of the spindle. (See also...) Figure 4 In this embodiment, the torque analysis algorithm can identify problems such as spindle imbalance, bending, and poor lubrication, and assess whether the spindle needs maintenance. It calculates the mean, standard deviation, and deviation of the torque based on torque data, sampling rate, and rated torque. The calculation formulas are as follows: in, The average torque. The number of sampling points. For the first One torque data point, For the standard deviation of torque, For the rated torque, the volatility is calculated based on the mean and standard deviation, and the trend of torque variation is analyzed. The calculation formulas are as follows: in, The regression slope represents the rate of change of torque over time. For the first One torque value, Using time indexing, when the calculated volatility exceeds a preset threshold, the periodicity of the torque signal fluctuation is analyzed. The condition for the existence of periodicity is: The health score of torque is calculated based on deviation, volatility, regression slope, and periodicity. The calculation formula is as follows: The condition of the spindle is determined based on the torque health score.

[0040] Strain analysis algorithms are used to analyze strain data and assess deformation in key areas. (See also...) Figure 5 In this embodiment, the strain analysis algorithm can identify problems such as structural over-limit and abnormal deformation trends, providing a basis for structural safety assessment. The safety factor and strain over-limit rate are calculated based on strain data, historical strain data, and strain limits, using the following formulas: in, For strain limits, To prepare for the current situation; Simultaneously, after analyzing the strain variation trend, the correlation between strains at different measuring points was analyzed, and the calculation formulas are as follows: in, The slope of the strain regression. For the first The time of the measurement For the first The strain value measured in this test. To measure the number of times, The Pearson correlation coefficient is used. For the measurement point A, the first This measurement value, For the measurement point B, the first This measurement value.

[0041] The strain health score is calculated based on the exceedance rate, trend judgment threshold, strain standard deviation, and strain standard deviation threshold. The calculation formula is as follows: If the mean is unavailable, the standard deviation can be used directly. The formula is: in, Points are awarded for exceeding the limit. This is the penalty coefficient for exceeding the limit. Punishment for adapting to trends The threshold for trend determination. To compensate for fluctuations, For strain standard deviation, The threshold for triggering fluctuations, The standard deviation threshold for strain. For the mean strain, For the number of times measured.

[0042] The strain over-limit penalty reflects the degree to which the current strain exceeds the safety limit; the strain trend penalty assesses the strain's change trend over time, identifying whether there is a risk of continuous increase or decrease in fatigue; and the strain fluctuation penalty assesses the degree of fluctuation in the strain signal, reflecting the impact of dynamic loads on the structure's fatigue. The corresponding strain state is graded based on the strain health score.

[0043] During fault detection, if the inner ring is faulty, it is recommended to replace the bearing; if the outer ring is faulty, it is recommended to check the bearing housing; if the rolling element is faulty, it is recommended to replace the rolling element; if the cage is faulty, it is recommended to replace the cage. If the torque fluctuation is too large, check the spindle imbalance and coupling alignment. If the trend continues to rise, check the load and bearing wear. If the trend continues to fall, check the transmission system. If the bearing temperature is high, check the lubrication and bearing damage. If the motor temperature is high, check the heat dissipation and load. If the temperature at the detection point is high, check for poor contact. If the spindle strain exceeds the limit, check the spindle deformation. If the lever strain exceeds the limit, check the lever deformation. If the strain at the connection exceeds the limit, check for loose connections.

[0044] A multi-dimensional state deviation algorithm based on weighted Euclidean distance is used for comprehensive judgment, generating a fault location report and handling strategy, which is displayed in real time through a visual interface, including: Define the current state vector and the normal baseline state vector; The overall deviation is calculated based on the current state vector, the normal baseline state vector, and the standard deviation of each feature under normal conditions. The calculation formula is as follows: Where n is the total dimension of the features, n=4. Let this be the current state vector. This is the normal baseline state vector. The standard deviation of each feature under normal conditions. For the weights of each dimension, This represents the overall deviation.

[0045] The final fault location report and handling strategy are generated based on the comprehensive deviation and displayed in real time through a visual interface.

[0046] Corresponding to the fault diagnosis method for a waterwheel-type loading and unloading robot provided in the above embodiments, this application also provides a fault diagnosis system for a waterwheel-type loading and unloading robot.

[0047] See Figure 6 This application provides a fault diagnosis system 20 for a waterwheel-type loading and unloading robot, comprising: The acquisition module 201 is used to acquire feature data of the waterwheel loading and unloading robot under various normal states, and to preprocess the feature data to form a multi-dimensional feature state set.

[0048] The determination module 202 is used to determine the threshold range of the waterwheel loading and unloading robot under various normal states based on the multi-dimensional feature state set.

[0049] The comparison and judgment module 203 is used to acquire the feature data of the water-wheel loading and unloading robot to be diagnosed and compare it with the threshold range, and judge the fault status of the water-wheel loading and unloading robot to be diagnosed based on the comparison result.

[0050] The fault diagnosis module 204 is used to analyze the data of the waterwheel loading and unloading robot that is determined to be in a fault state, and then use a multi-dimensional state deviation algorithm based on weighted Euclidean distance to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

[0051] See Figure 7In this embodiment, to construct a highly specialized and scalable intelligent diagnostic decision-making system, this system adopts a multi-agent collaborative analysis architecture based on the LangGraph framework. Vibration frequency domain analysis tools, torque period similarity analysis tools, strain rainflow counting tools, and temperature trend analysis tools are encapsulated and registered in the LangGraph tool library. Through an access control mechanism, only the corresponding expert agents are authorized to use these tools, ensuring the security and professionalism of tool usage. This architecture uses a large model agent as the core collaborative control center, responsible for understanding the global diagnostic task, performing high-order reasoning, and formulating analysis plans. During execution, this control center dynamically distributes subtasks to multiple expert agents with specific domain knowledge, such as vibration analysis experts, thermodynamic analysis experts, and structural mechanics analysis experts, based on fault symptoms. Each expert agent is authorized to independently call its own dedicated, encapsulated analysis toolchain, such as FFT analysis, steady-state temperature calculation, and rainflow counting, to perform in-depth mining and feature interpretation of the corresponding sensor data streams and form preliminary local diagnostic conclusions. The diagnostic interface highlights abnormal parameters and displays intelligent diagnostic conclusions in a pop-up window, such as: abnormal vibration second harmonic energy, deviation 0.85, indicating bearing misalignment, and recommends stopping the machine for calibration. The interface provides a detailed fault analysis report, including fault type, fault location, fault cause, fault severity, confidence level, and recommended measures. The diagnostic process data, tool call records, and final conclusions are automatically stored in the historical database as training data for subsequent correction of the baseline state vector and optimization weights. The database uses relational databases such as MySQL and PostgreSQL to store structured data, time-series databases such as InfluxDB to store sensor time-series data, and document databases such as MongoDB to store diagnostic reports and unstructured data. After report generation and storage, the system status is reset, and the next round of data acquisition and status monitoring continues, realizing a complete closed loop of monitoring-alarm-diagnosis-decision-optimization-re-monitoring. Through continuous learning and optimization, the system continuously improves diagnostic accuracy and predictive capabilities, providing reliable assurance for the safe and stable operation of the water-wheel loading and unloading robot.

[0052] In this embodiment, "multiple" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent the existence of A alone, the simultaneous existence of A and B, or the existence of B alone. A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0053] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0054] The above description is merely a specific embodiment of this application. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the protection scope of this application. The protection scope of this application should be determined by the protection scope of the claims.

Claims

1. A fault diagnosis method for a waterwheel-type loading and unloading robot, characterized in that, include: The feature data of the water-wheel loading and unloading robot under various normal states are acquired, and the feature data is preprocessed to form a multi-dimensional feature state set. The threshold range of the waterwheel loading and unloading robot under each normal state is determined based on the multidimensional feature state set. After acquiring the feature data of the water-wheel loading and unloading robot to be diagnosed, it is compared with the threshold range, and the fault status of the water-wheel loading and unloading robot to be diagnosed is determined based on the comparison results. After analyzing the data of the waterwheel loading and unloading robot that is determined to be in a faulty state, a multi-dimensional state deviation algorithm based on weighted Euclidean distance is used to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

2. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 1, characterized in that, The process involves acquiring feature data of the waterwheel-type loading and unloading robot under various normal states, and preprocessing the feature data to form a multi-dimensional feature state set, including: Vibration sensors are used to collect the raw three-axis acceleration time series signals of the water-wheel loading and unloading robot at a high frequency sampling rate. After filtering out noise through a digital filter, the noise-reduced signal is standardized. The time-domain features are calculated from the standardized time-series original signal, and the signal is transformed to the frequency domain by fast Fourier transform and the frequency domain features are calculated. The temporal and frequency domain features are combined to form a vibration feature vector; Torque signals are collected during the operation of the water-wheel loading and unloading robot by a torque sensor. The complete working cycle is divided by key point detection to obtain a periodic torque sequence. The dynamic time warping algorithm is used to align multiple cycles in time and calculate the statistical and shape features in each aligned cycle. The statistical and shape features are then combined to form a torque feature vector. Temperature data of key parts of the water-wheel loading and unloading robot are collected by temperature sensors, and the trend signal is obtained by smoothing with a moving average filter. The standard deviation within the sliding window is then calculated. The standard deviation is used to determine whether the water-wheel loading and unloading robot has entered the thermal steady-state working stage; The average value within the steady-state window is calculated as the core temperature variable, and the temperature rise rate from startup to steady state is calculated. The core temperature variable and the temperature rise rate are combined to form a temperature feature vector. The strain signals of key parts of the waterwheel loading and unloading robot are collected by strain sensors, obvious abnormal jump points are eliminated, and the collected voltage signals are converted into micro-strain physical quantities according to the bridge formula and sensor sensitivity coefficient. The average strain and maximum strain during the monitoring period are calculated. The amplitude-mean spectrum of strain fluctuations is statistically analyzed using the rainflow counting method to obtain the fatigue damage degree. The average strain, maximum strain, and fatigue damage are combined to form a strain feature vector; The vibration feature vector, torque feature vector, temperature feature vector, and strain feature vector are aligned and fused within a unified time window to form a multidimensional feature state set.

3. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 1, characterized in that, The threshold ranges for the waterwheel-type loading and unloading robot under various normal states are determined based on the multidimensional feature state set, including: The mean and standard deviation of each feature parameter in the multidimensional feature state set are calculated using statistical analysis methods. The threshold range is set based on the calculated mean and standard deviation, and the threshold range is dynamically adjusted according to the operating conditions of the waterwheel loading and unloading robot.

4. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 1, characterized in that, The process of acquiring the feature data of the water-wheel loading and unloading robot to be diagnosed and comparing it with the threshold range, and then determining the fault status of the water-wheel loading and unloading robot to be diagnosed based on the comparison result, includes: Obtain the current operating status of the waterwheel loading and unloading robot, and select the corresponding threshold range based on the current status; After determining the current state, the vibration data, temperature data, torque data, and strain data of the water-wheel loading and unloading robot to be diagnosed are obtained respectively. The vibration data, temperature data, torque data, and strain data are preprocessed to obtain the feature vector of the current state; The feature vector of the current state is compared with the corresponding threshold range to determine whether each feature parameter is within the threshold range. If all feature parameters are within the threshold range, the waterwheel loading and unloading robot is determined to be in normal condition. If any feature parameter exceeds the threshold range, the waterwheel loading and unloading robot is determined to be in a fault state. If multiple characteristic parameters exceed the threshold range simultaneously, the waterwheel loading and unloading robot is determined to be in a serious malfunction state.

5. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 1, characterized in that, The analysis of data from the water-wheel loading and unloading robot determined to be in a faulty state includes: Real-time acquisition of vibration, temperature, torque, and strain data under fault conditions; The vibration data is analyzed using a vibration analysis algorithm to diagnose the health status of the bearing. The temperature data is analyzed using a temperature analysis algorithm to identify abnormal operating conditions and determine the abnormal measurement points and causes. The torque data is analyzed using a torque analysis algorithm to diagnose the condition of the spindle; The strain data is analyzed using a strain analysis algorithm to assess the deformation of key components.

6. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 5, characterized in that, The step of analyzing the vibration data using a vibration analysis algorithm to diagnose the health status of the bearing includes: Based on the vibration data, sampling rate, and bearing parameters, the cage frequency, inner ring failure frequency, outer ring failure frequency, rolling element failure frequency, and root mean square value are calculated respectively, using the following formulas: in, For frequency conversion, The shaft speed is To maintain the rack frequency, The diameter of the rolling element, The diameter of the pitch circle. Contact angle, For the inner ring fault frequency, For the number of rolling elements, For the outer ring fault frequency, For the rolling element failure frequency, The number of sampling points. For the first One vibration acceleration data point, It is the root mean square value; The frequency matching range is determined based on the calculated theoretical bearing fault characteristic frequencies. The calculation formula is as follows: in, To match the frequency range, This represents the theoretical characteristic frequency of bearing failure. The vibration fault is determined based on the matched frequency range combined with the detected peak frequency, peak amplitude, and amplitude detection threshold. The calculation formula is as follows: in, The detected peak frequency, The detected peak amplitude, The amplitude detection threshold; The vibration RMS penalty is determined based on the root mean square value, and the penalty fault is determined based on the detected fault frequency. The vibration health score is calculated based on the vibration RMS penalty and the penalty fault, using the following formula: in, The peak amplitude corresponding to the detected fault frequency; The remaining life of the vibration bearing is estimated based on the vibration health score.

7. The fault diagnosis method for water-wheel type loading and unloading robot according to claim 5, characterized in that, The step of analyzing the torque data using a torque analysis algorithm to diagnose the spindle's condition includes: The mean, standard deviation, and deviation of the torque are calculated based on the torque data, sampling rate, and rated torque, using the following formulas: in, The average torque. The number of sampling points. For the first One torque data point, For the standard deviation of torque, This is the rated torque; Volatility is calculated based on the mean and standard deviation, and the trend of torque variation is analyzed. The calculation formulas are as follows: in, The regression slope, For the first One torque value, For time indexing; When the calculated volatility exceeds a preset threshold, the periodicity of the torque signal fluctuation is analyzed. The health score of torque is calculated based on the aforementioned deviation, volatility, regression slope, and periodicity, using the following formula: The condition of the spindle is determined based on the torque health score.

8. The fault diagnosis method for the water-wheel loading and unloading robot according to claim 5, characterized in that, The step of analyzing the strain data using a strain analysis algorithm to evaluate the deformation of key components includes: The safety factor and strain exceedance rate are calculated based on the strain data, strain history data, and strain limits, respectively, using the following formulas: in, For strain limits, To prepare for the current situation; Simultaneously, after analyzing the strain variation trend, the correlation between strains at different measuring points was analyzed, and the calculation formulas are as follows: in, The slope of the strain regression. For the first The time of the measurement For the first The strain value measured in this test. To measure the number of times, The Pearson correlation coefficient is used. For the measurement point A, the first This measurement value, For the measurement point B, the first The measurement value; The strain health score is calculated based on the over-limit rate, trend judgment threshold, strain standard deviation, and strain standard deviation threshold. The calculation formula is as follows: in, Points are awarded for exceeding the limit. This is the penalty coefficient for exceeding the limit. Punishment for adapting to trends The threshold for trend determination To compensate for fluctuations, For strain standard deviation, The threshold for triggering fluctuations, The standard deviation threshold for strain. For the mean strain, For the number of measurements; The corresponding strain state is classified according to the strain health score.

9. The fault diagnosis method for the waterwheel-type loading and unloading robot according to claim 1, characterized in that, A multi-dimensional state deviation algorithm based on weighted Euclidean distance is used for comprehensive judgment, generating a fault location report and handling strategy, which is displayed in real time through a visual interface, including: Define the current state vector and the normal baseline state vector; The overall deviation is calculated based on the current state vector, the normal baseline state vector, and the standard deviation of each feature under normal conditions. The calculation formula is as follows: Where n is the total dimension of the features. Let this be the current state vector. This is the normal baseline state vector. The standard deviation of each feature under normal conditions. For the weights of each dimension, This refers to the overall deviation. Based on the comprehensive deviation, a final fault location report and handling strategy are generated and displayed in real time through a visual interface.

10. A fault diagnosis system for a waterwheel-type loading and unloading robot, characterized in that, include: The acquisition module is used to acquire feature data of the waterwheel loading and unloading robot under various normal states, and to preprocess the feature data to form a multi-dimensional feature state set. The determination module is used to determine the threshold range of the waterwheel loading and unloading robot under each normal state based on the multidimensional feature state set; The comparison and judgment module is used to acquire the feature data of the water-wheel loading and unloading robot to be diagnosed and compare it with the threshold range, and judge the fault status of the water-wheel loading and unloading robot to be diagnosed based on the comparison result. The fault diagnosis module analyzes the data of the waterwheel loading and unloading robot that is determined to be in a faulty state, and then uses a multi-dimensional state deviation algorithm based on weighted Euclidean distance to make a comprehensive judgment, generate a fault location report and processing strategy, and display it in real time through a visual interface.

Images