Humanoid robot reliability test method and system

Abstract

The application provides a humanoid robot reliability test method and system, and belongs to the technical field of humanoid robot testing. In view of the problems of single test scene, one-sided dimension, backward judgment, low efficiency and lack of life prediction, the application constructs a comprehensive test scene library integrating physical environment and task scene, generates high-coverage test cases by using orthogonal test method, automatically collects data through multi-sensor fusion, identifies failure modes after processing, establishes a performance degradation model, and finally realizes reliability quantitative evaluation and residual life prediction of key components and the whole machine. Through the construction of the comprehensive test scene library and the use of the orthogonal test method, the comprehensiveness and efficiency of the test are realized. Through the automatic collection of data by multi-sensor fusion, the automation of the test is realized. The application provides a basis for preventive maintenance and effectively supports the reliability improvement and large-scale commercial application of humanoid robots.

Description

Technical Field

[0001] This invention relates to the field of humanoid robot testing technology, and specifically to a method and system for testing the reliability of humanoid robots. Background Technology

[0002] Humanoid robots, with their anthropomorphic form, have broad application prospects in various fields such as service industry, medical rehabilitation, hazardous environment operation, and family companionship. However, their complex mechanical structure, numerous drive systems, highly integrated perception and control systems, and highly dynamic motion characteristics pose serious challenges to their reliability. Insufficient reliability can lead to frequent robot failures, increased maintenance costs, deteriorated user experience, and even personal safety accidents, severely hindering their large-scale commercialization and practical application.

[0003] Currently, industrial robots have relatively mature reliability testing and evaluation standards. These standards mainly assess the static or quasi-static performance indicators of robots in structured and deterministic environments. However, humanoid robots operate in open, complex, and unstructured environments, and their tasks are highly dynamic and interactive. Directly applying industrial robot testing methods has many shortcomings. Testing scenarios are limited to flat, well-organized laboratory environments, failing to simulate the diverse ground materials, slopes, temperatures, humidity levels, and dynamic obstacles encountered in the real world, making it difficult to expose potential environmental adaptability deficiencies. Testing dimensions are one-sided, focusing only on single performance indicators and neglecting the interactions between subsystems such as mechanics, drive, control, and energy, thus failing to provide comprehensive data support for overall reliability optimization. Failure determination methods are outdated, relying on manual observation and simple threshold judgments, lacking objective quantitative standards, making it difficult to identify gradual performance degradation, predict early failures, increase maintenance costs, and reduce production efficiency. Testing efficiency is low, automation is minimal, requiring numerous repetitive tests with long cycles and high costs, hindering rapid product iteration. The lack of lifespan prediction capabilities, limiting testing to "pass / fail" verification tests, leads to passive maintenance strategies and fails to meet the needs of long-term, unattended applications. Summary of the Invention

[0004] The present invention aims to solve the problems mentioned in the background art by providing a method and system for reliability testing of humanoid robots.

[0005] The specific technical solution is as follows:

[0006] A humanoid robot reliability testing system includes a central control unit and a test scenario construction unit, a test case generation unit, an automated test execution unit, and a reliability assessment and lifespan prediction unit, all of which are communicatively connected to the central control unit.

[0007] The central control unit is used to schedule the collaborative work of various units to achieve unified control of the testing process; the test scenario construction unit is used to build a comprehensive test scenario library covering various actual working conditions; the test case generation unit is used to generate a high-coverage test case set based on the comprehensive test scenario library; the automated test execution unit is used to drive the humanoid robot to execute test cases and collect various data during the robot's operation in real time; the data processing and analysis unit is used to process the collected data, identify failure modes, and analyze performance degradation trends; the reliability assessment and life prediction unit is used to conduct a quantitative reliability assessment based on the analysis results and predict the remaining service life of the robot and key components.

[0008] The aforementioned humanoid robot reliability testing system includes a test scenario construction unit comprising a physical environment simulation module and a task scenario definition module. The physical environment simulation module is used to simulate physical environments with different temperatures and humidity, ground materials, slopes, and obstacle distributions. The task scenario definition module is used to define task types such as walking, climbing stairs, grasping and placing, and turning in narrow passages. The physical environment simulation module and the task scenario definition module work together to form a comprehensive test scenario library that includes combinations of environment and task.

[0009] In the aforementioned humanoid robot reliability testing system, the test case generation unit designs test cases using orthogonal experimental design. The test case generation unit obtains multiple factors affecting the reliability of the humanoid robot and the different levels of each factor, selects the corresponding orthogonal array, extracts working condition combinations from the comprehensive test scenario library, and generates a test case set covering the main influencing factors.

[0010] The aforementioned humanoid robot reliability testing system includes an automated test execution unit comprising a multi-sensor fusion module and a motion control module. The multi-sensor fusion module is deployed on key components of the humanoid robot, such as joints, motors, power supplies, and sensing components, and is used to collect data such as joint angles, torque, temperature, current and voltage, vibration, and visual processing delay. The motion control module is used to receive instructions from the central control unit and drive the humanoid robot to perform corresponding actions according to the test cases.

[0011] The aforementioned humanoid robot reliability testing system includes a data processing and analysis unit comprising a data preprocessing module and a degradation analysis module. The data preprocessing module is used to clean, filter, and reduce noise in the collected raw data. The degradation analysis module is used to determine whether key parameters in the data exceed preset thresholds, identify and record sudden failure modes, and establish a performance degradation trajectory model for parameters that change gradually.

[0012] The aforementioned humanoid robot reliability testing system includes a reliability assessment and life prediction unit comprising a reliability index calculation module and a life prediction module. The reliability index calculation module is used to calculate reliability indices such as mean time between failures (MTBF) and reliability based on failure mode records and performance degradation data. The life prediction module is used to extrapolate the performance degradation trajectory to a preset failure threshold to obtain the remaining service life of the robot and its key components.

[0013] In the aforementioned humanoid robot reliability testing system, the central control unit adopts a centralized control architecture, establishes data interaction with each unit through a preset communication protocol, and realizes intelligent decision-making for test parameter configuration, test process scheduling, and abnormal situation response, as well as adaptive adjustment of test strategies based on the robot's real-time operating status.

[0014] The present invention also provides a method for testing the reliability of a humanoid robot based on the above-described humanoid robot reliability testing system, comprising the following steps:

[0015] Step 1: Build a comprehensive test scenario library, integrating physical environments and task scenarios to form a set of test scenarios covering various actual working conditions;

[0016] Step 2: Based on the comprehensive test scenario library, use experimental design methods to generate a high-coverage test case set;

[0017] Step 3: Automated drive the humanoid robot to execute the test case set, synchronously collecting various status data and performance data during the robot's operation;

[0018] Step 4: Process the collected data, identify failure modes, and establish a performance degradation trajectory model; Step 5: Based on the failure modes and performance degradation trajectory model, conduct a quantitative reliability assessment and predict the remaining service life of the robot and key components.

[0019] The aforementioned humanoid robot reliability testing method, wherein step one, the process of constructing a comprehensive test scenario library, includes: constructing a physical environment library covering low temperature, normal temperature, high temperature, low humidity, normal humidity, high humidity, smooth ground, rough ground, gravel ground, flat ground, slope, and stairs; defining a task type library including continuous walking, repeated climbing up and down stairs, load-bearing grasping and placement, maintaining balance under disturbance, and turning in narrow passages; and combining the environmental conditions in the physical environment library with the task types in the task type library to form a comprehensive test scenario library.

[0020] The aforementioned humanoid robot reliability testing method includes the following steps: Step 4, establishing a performance degradation trajectory model, involves: extracting features from the preprocessed data, including time-domain features, frequency-domain features, or time-frequency-domain features; determining whether key parameters exceed preset thresholds, and if so, marking it as a sudden failure and recording the failure mode; if the parameters do not exceed the thresholds but exhibit gradual changes, establishing a linear or exponential performance degradation trajectory model based on the extracted features; and Step 5, the lifetime prediction process, using neural network or support vector machine algorithms to train and optimize the performance degradation trajectory model to improve the accuracy of remaining lifetime prediction.

[0021] The present invention has the following beneficial effects:

[0022] In terms of test comprehensiveness, the integrated test scenario library incorporates diverse physical environments and task types, simulating real and complex working conditions. This effectively exposes potential defects in the robot's environmental adaptability and multi-subsystem collaborative operation, significantly improving the realism and coverage of the tests. Regarding test efficiency, the application of orthogonal experimental design significantly reduces the number of test cases. Combined with automated test execution processes, it avoids redundancy and inefficiency of manual operations, shortens the test cycle, reduces test costs, and meets the needs of rapid product iteration. In terms of failure identification and assessment, multi-sensor fusion enables comprehensive acquisition of multi-dimensional data, and data preprocessing and feature extraction are accurate and efficient. By ensuring data quality, performance degradation modeling can accurately capture sudden failures and gradual performance degradation, replacing the traditional judgment method that relies on human experience, and realizing an objective quantitative assessment of reliability. In terms of maintenance and application value, the life prediction function based on the degradation model breaks the passive situation of "post-event maintenance" or fixed-cycle maintenance, provides a scientific basis for preventive maintenance, and reduces long-term application risks. At the same time, the centralized control architecture and intelligent adaptive control strategy ensure the stable and coordinated operation of the system. The overall solution has universality and scalability, can be adapted to different types of robots, and provides complete technical support for improving robot reliability. Attached Figure Description

[0023] Figure 1 A connection diagram of a humanoid robot reliability testing system provided in an embodiment of the present invention;

[0024] Figure 2 A flowchart of a humanoid robot reliability testing method provided in an embodiment of the present invention. Detailed Implementation

[0025] The technical solution of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0026] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual images. They should not be construed as limiting the scope of this application. To better illustrate the embodiments of the present invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual dimensions of the product. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0027] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "inner," and "outer" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present application. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0028] In the description of this invention, unless otherwise explicitly specified and limited, the term "connection" or similar designation indicating a connection between components should be interpreted broadly. For example, it can refer to a fixed connection, a detachable connection, or an integral part; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediate medium; it can refer to the internal communication between two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0029] Example 1

[0030] See attached document Figure 1 This embodiment provides a humanoid robot reliability testing system, including a central control unit and a test scenario construction unit, a test case generation unit, an automated test execution unit, a data processing and analysis unit, and a reliability assessment and lifespan prediction unit, all of which are communicatively connected to the central control unit.

[0031] The central control unit is used to schedule the collaborative work of various units and achieve unified control of the testing process; the test scenario construction unit is used to build a comprehensive test scenario library covering various actual working conditions; the test case generation unit is used to generate a high-coverage test case set based on the comprehensive test scenario library; the automated test execution unit is used to drive the humanoid robot to execute test cases and collect various data during the robot's operation in real time; the data processing and analysis unit is used to process the collected data, identify failure modes and analyze performance degradation trends; the reliability assessment and life prediction unit is used to conduct quantitative reliability assessment based on the analysis results and predict the remaining service life of the robot and key components.

[0032] By constructing a multi-unit collaborative system architecture and leveraging a central control unit to achieve unified control over the entire process, the system integrates functions such as scenario building, test case generation, automated execution, data processing, and evaluation and prediction. This breaks through the fragmented limitations of existing testing methods, realizes the systematization and automation of humanoid robot reliability testing, comprehensively covers the performance of the entire robot system, provides complete data support for reliability assessment and life prediction, and solves the problem of the lack of systematic testing solutions in existing technologies.

[0033] Specifically, in this embodiment, the test scenario construction unit includes a physical environment simulation module and a task scenario definition module. The physical environment simulation module simulates physical environments with different temperatures, humidity levels, ground materials, slopes, and obstacle distributions. The task scenario definition module defines task types such as walking, climbing stairs, grasping and placing, and turning in narrow passages. The physical environment simulation module and the task scenario definition module work together to form a comprehensive test scenario library that includes combinations of environments and tasks. By combining the physical environment simulation module and the task scenario definition module, diverse physical environments and various task types are combined to construct a comprehensive test scenario library that closely resembles real-world applications. This effectively solves the shortcomings of existing test scenarios that are too singular and unable to simulate real working conditions. It can fully expose the robot's potential adaptability issues under different environments and tasks, improving the realism and comprehensiveness of the test.

[0034] Specifically, in this embodiment, the test case generation unit uses orthogonal experimental design to design test cases. The unit acquires multiple factors affecting the reliability of the humanoid robot and their different levels, selects corresponding orthogonal arrays, extracts working condition combinations from a comprehensive test scenario library, and generates a test case set covering the main influencing factors. By using orthogonal experimental design to select representative test cases from multi-factor, multi-level working condition combinations, redundancy in comprehensive testing is avoided. While ensuring test coverage, the number of tests is significantly reduced, solving the problems of low efficiency, long cycles, and high costs in existing testing methods. Furthermore, by clarifying the primary and secondary influences of factors, targeted directions are provided for test optimization and design improvement, enhancing the scientific rigor of the testing.

[0035] Specifically, in this embodiment, the automated test execution unit includes a multi-sensor fusion module and a motion control module. The multi-sensor fusion module is deployed on key parts of the humanoid robot, such as joints, motors, power supplies, and sensing components, to collect data on joint angles, torque, temperature, current and voltage, vibration, and visual processing latency. The motion control module receives instructions from the central control unit and drives the humanoid robot to perform corresponding actions according to the test cases. By deploying multiple types of sensors on key parts of the robot, real-time acquisition of multi-dimensional performance data of core components such as joints, motors, and power supplies is achieved. Combined with the automated drive execution of the motion control module, this overcomes the shortcomings of existing tests that are one-dimensional and reliant on manual operation, ensuring the comprehensiveness, real-time nature, and accuracy of data acquisition, and providing rich and reliable raw data for subsequent performance analysis.

[0036] Specifically, in this embodiment, the data processing and analysis unit includes a data preprocessing module and a degradation analysis module. The data preprocessing module is used to clean, filter, and reduce noise in the collected raw data. The degradation analysis module is used to determine whether key parameters in the data exceed preset thresholds, identify and record sudden failure modes, and establish performance degradation trajectory models for parameters that show gradual changes. By purifying the raw data through the data preprocessing module to remove interference information, and then having the degradation analysis module identify sudden failure modes and establish performance degradation trajectory models, this approach changes the outdated method of relying on subjective observation and simple threshold judgment in existing testing. It can objectively and accurately capture sudden failures and gradual performance degradation of the robot, providing high-quality analytical data for reliability assessment and solving the problem of being unable to identify potential performance degradation.

[0037] Specifically, in this embodiment, the reliability assessment and lifespan prediction unit includes a reliability index calculation module and a lifespan prediction module. The reliability index calculation module is used to calculate reliability indices such as mean time between failures (MTBF) and reliability based on failure mode records and performance degradation data. The lifespan prediction module is used to extrapolate the performance degradation trajectory to a preset failure threshold to obtain the remaining service life of the robot and its key components. By calculating quantitative indices such as MTBF and reliability, an objective assessment of reliability is achieved. Furthermore, by extrapolating the performance degradation trajectory to predict the remaining service life, this overcomes the limitation of existing technologies that can only perform "pass / fail" verification. It solves the problems of lacking lifespan prediction capabilities and passive maintenance strategies, providing a scientific basis for robot design optimization, fault prevention, and preventative maintenance.

[0038] Specifically, in this embodiment, the central control unit adopts a centralized control architecture. It establishes data interaction with each unit through a preset communication protocol, enabling intelligent decision-making for test parameter configuration, test process scheduling, and abnormal situation response, as well as adaptive adjustment of test strategies based on the robot's real-time operating status. The centralized control architecture achieves unified scheduling and data interaction among the units. Combined with intelligent decision-making and adaptive control strategies, it can adjust test parameters and processes according to the robot's real-time operating status, avoiding rigid control during the testing process and ensuring stable and collaborative operation of the testing system under complex working conditions. It also improves the test's responsiveness to the robot's real-time status, further optimizing test efficiency and quality.

[0039] Specifically, in this embodiment, the lifespan prediction module uses a comprehensive performance degradation evaluation equation to quantify the degradation state of the robot as a whole or key components. The equation is as follows:

[0040] ;

[0041] in:

[0042] D(t) represents the performance degradation exponent at time t. The larger the value, the more severe the degradation. The example value is 0.85.

[0043] This represents the measured value of the i-th key performance parameter at time t, such as joint torque, motor temperature, vibration amplitude, etc.

[0044] This indicates the initial health status value of the corresponding parameter, which is the factory calibration value;

[0045] This indicates the cumulative task execution time; the example value is 120h.

[0046] This indicates the design lifespan of the task; the example value is 1000 hours.

[0047] This represents the real-time intensity of the j-th type of environmental factor (such as temperature, humidity, and ground unevenness).

[0048] This represents the design threshold for the corresponding environmental factor;

[0049] α, β, γ are weighting coefficients, obtained through training with historical data;

[0050] These are the environmental factor weights, reflecting the degree of impact of different environments on degradation;

[0051] n and m represent the number of performance parameters and environmental factors, respectively.

[0052] Equation derivation process:

[0053] This equation integrates the effects of performance parameter deviation, cumulative effect of mission time, and coupling effect of environmental stress:

[0054] Part 1:

[0055] ;

[0056] This represents the normalized mean square deviation of each performance parameter relative to the initial state, reflecting the consistent degradation of the overall system performance.

[0057] 2. Part Two:

[0058] ;

[0059] A decay function for the cumulative task time is introduced to simulate the nonlinear effect of decreased system tolerance as task execution time increases.

[0060] 3. Part Three:

[0061] ;

[0062] The real-time impact of multiple environmental factors is quantified, and the contribution of different environmental factors is adjusted by weighting.

[0063] By combining the above three parts with linear weighting, a comprehensive degradation index D(t) is formed, which takes into account both internal performance changes and the coupling effect of external tasks and environment.

[0064] Example:

[0065] Suppose we are conducting a degradation assessment on the leg joint system of a humanoid robot:

[0066] Select the following performance parameters: joint torque error S1, motor temperature rise S2, and vibration peak value S3;

[0067] Environmental factors: surface roughness F1, ambient temperature F2;

[0068] Initial values: S1(0) = 0.1 Nm, S2(0) = 30°C, S3(0) = 0.5 m / s²;

[0069] Threshold: =2.0 (roughness coefficient) =50°C;

[0070] Weight: α=0.6, β=0.4, γ=1.2, w1=0.7, w2=0.3.

[0071] Real-time data is collected and then substituted into the equation to calculate D(t). If D(t) exceeds a preset threshold (such as 1.0), an early warning or lifespan termination judgment is triggered.

[0072] Technical effects:

[0073] 1. Multi-source information fusion: Integrating internal performance data with external environmental data to improve the comprehensiveness of the assessment;

[0074] 2. Nonlinear modeling: The introduction of an exponential decay function better reflects the accelerated performance degradation phenomenon in actual engineering.

[0075] 3. High interpretability: The physical meaning of each part is clear, facilitating debugging and optimization;

[0076] 4. Supports adaptive prediction: Weights can be updated in real time to adapt to individual differences among robots;

[0077] 5. Improved lifetime prediction accuracy: Compared with the single-parameter threshold method, this equation provides a smoother and more realistic degradation trajectory.

[0078] Working principle and process:

[0079] 1. Data Acquisition Phase: Data is acquired through a multi-sensor fusion module. and ;

[0080] 2. Data processing stage: Calculate the ratio of each deviation to the ratio of environmental factors;

[0081] 3. Degradation index calculation: Substitute into the equation to obtain D(t);

[0082] 4. Lifetime prediction stage: When D(t) reaches the preset failure threshold When the lifespan is deemed to have ended, the remaining lifespan is... ;

[0083] 5. Feedback optimization: Iteratively update the weight coefficients based on measured degradation data to improve model adaptability.

[0084] Example 2

[0085] See attached document Figure 2 This embodiment provides a humanoid robot reliability testing method based on the humanoid robot reliability testing system of Embodiment 1, including the following steps:

[0086] Step 1: Build a comprehensive test scenario library, integrating physical environments and task scenarios to form a set of test scenarios covering various actual working conditions;

[0087] Step 2: Based on a comprehensive test scenario library, use experimental design methods to generate a high-coverage test case set;

[0088] Step 3: Automated humanoid robot executes test case set, synchronously collecting various status data and performance data during robot operation;

[0089] Step 4: Process the collected data, identify failure modes, and establish a performance degradation trajectory model; Step 5: Based on the failure modes and performance degradation trajectory model, conduct a quantitative reliability assessment and predict the remaining service life of the robot and key components.

[0090] By establishing a complete process of scenario construction, test case generation, automatic execution, data processing, and evaluation and prediction, a data-driven testing closed loop is formed. This systematically solves the shortcomings of existing testing methods in terms of scenario realism, dimensional comprehensiveness, judgment objectivity, process efficiency, and result predictability. It realizes the scientific and automated full-process reliability testing of humanoid robots, providing complete methodological support for improving robot reliability.

[0091] Specifically, in this embodiment, the process of constructing the comprehensive test scenario library in step one includes: constructing a physical environment library covering low temperature, normal temperature, high temperature, low humidity, normal humidity, high humidity, smooth ground, rough ground, gravel ground, flat ground, slope, and stairs; defining a task type library including continuous walking, repeated climbing and descending stairs, weighted grasping and placement, maintaining balance under disturbance, and turning in narrow passages; and combining the environmental conditions in the physical environment library with the task types in the task type library to form the comprehensive test scenario library. By subdividing and combining physical environment types and task types, the coverage of the comprehensive test scenario library is further refined, making the test scenarios more closely resemble the complex working conditions in actual robot applications. This allows for more targeted exposure of potential defects in robots under different environmental conditions and task requirements, improving the representativeness and comprehensiveness of the test scenarios and laying a solid foundation for the subsequent generation of high-coverage test cases.

[0092] Specifically, in this embodiment, the process of establishing a performance degradation trajectory model in step four includes: extracting features from the preprocessed data, including time-domain features, frequency-domain features, or time-frequency-domain features; determining whether key parameters exceed preset thresholds, and if so, marking it as a sudden failure and recording the failure mode; if the parameters do not exceed the thresholds but show a gradual change, then establishing a linear or exponential performance degradation trajectory model based on the extracted features; in step five, the lifespan prediction process uses neural network or support vector machine algorithms to train and optimize the performance degradation trajectory model, improving the accuracy of remaining lifespan prediction. By extracting multi-dimensional features to establish a performance degradation model and using machine learning algorithms to optimize the model, the accuracy of performance degradation identification and the reliability of lifespan prediction are improved, solving the problems of insufficient adaptability and limited prediction accuracy of a single model. At the same time, performance degradation analysis and lifespan prediction are more closely aligned with the actual operating state of the robot, providing more accurate and reliable support for reliability assessment and maintenance decisions.

[0093] Specifically, in this embodiment, the specific schemes for the neural network and support vector machine algorithms in the lifetime prediction module are as follows:

[0094] I. The specific data input and preprocessing specifications are as follows:

[0095] The input data contains two types of core features, totaling n+m dimensions (n ​​being the number of performance parameters and m being the number of environmental factors). Performance degradation features are time-domain features (mean, peak value, variance, root mean square), frequency-domain features (power spectral density, center frequency), and time-frequency-domain features (wavelet packet energy entropy) extracted from the preprocessed data, corresponding to n performance parameters (such as joint torque error, motor temperature rise, vibration peak value, etc.); environmental coupling features are real-time environmental factor data (such as ground roughness, ambient temperature, humidity, etc.), corresponding to m environmental factors.

[0096] The preprocessing steps are as follows:

[0097] 1. Standardization processing uses the StandardScaler method to normalize the input features. The formula is Xscaled=(X-μ) / σ, where μ is the feature mean, σ is the feature standard deviation, Xscaled is the standardized feature data, and X represents the original feature data before standardization, i.e., the single feature sample value or feature vector to be used in model training, thereby eliminating the influence of dimensional differences on model training.

[0098] 2. Randomly divide the training set and the test set in an 8:2 ratio. The training set is used for model parameter fitting, and the test set is used for model performance evaluation.

[0099] 3. When the feature dimension n+m≥10, principal component analysis (PCA) is used to reduce the feature dimension to 2-5 dimensions, retaining principal components with a cumulative variance contribution rate ≥95%, thus reducing the computational load of the model.

[0100] II. The specific implementation scheme of the Support Vector Machine (SVM) algorithm is as follows:

[0101] A regression-based support vector machine (SVR) was used to fit the nonlinear mapping relationship between performance degradation characteristics and remaining lifetime.

[0102] The kernel function chosen is the RBF (Radial Basis Function), with the formula K(x_i,x_j)=exp(-γ||x_i-x_j||²), which adapts to the nonlinear degradation under complex conditions and has better generalization ability than linear and polynomial kernels. The penalty coefficient C ranges from 1.0 to 10.0, with a default value of 1.0. It can be increased to 5.0-10.0 when the training set is overfitted, and decreased to 0.5-1.0 when the error on the test set is too large, to balance the model's fitting accuracy and generalization ability. The kernel function parameter γ is set to scale or 0.01-0.1. In scale mode, γ is automatically calculated as 1 / (n_features*X.var()), and when set manually, 0.05 is preferred to control the local influence range of the kernel function. The loss function adopts ε-insensitive loss, with an error tolerance ε of 0.1-0.2, allowing for a small range of prediction errors and improving model robustness.

[0103] Where K(x_i,x_j) is the kernel function output value, which is the inner product of two sample vectors after being mapped to a high-dimensional feature space, and is used to measure the similarity between samples;

[0104] x_i is the feature vector of the i-th sample, the standardized feature set of a single training or test sample, with dimensions n+m (performance parameters + environmental factors).

[0105] x_j is the feature vector of the j-th sample, and the other feature vector of the sample corresponding to x_i is used in the same way as x_i;

[0106] γ is a kernel function parameter that controls the local influence range of the RBF kernel and determines the sensitivity of the model to local features of the sample.

[0107] ||x_i-x_j||² represents the squared Euclidean distance between the feature vectors of two samples, quantifying the degree of difference between the two samples in the original feature space.

[0108] The training and optimization process is as follows:

[0109] 1. Based on the above parameters, construct the SVR model. You can call the sklearn.svm.SVR interface or equivalent open-source tools or self-developed code to complete the initialization.

[0110] 2. Input the standardized training set into the model, and solve for the optimal separating hyperplane using the Sequence Minimum Optimization (SMO) algorithm to obtain the initial model.

[0111] 3. Calculate the mean squared error (MSE) using the test set. The formula is MSE=Σ(y_true-y_pred)² / N (where N is the number of test set samples, the total number of test or validation set samples used in the error calculation). The target MSE is ≤0.05. y_true is the true remaining lifetime value, representing the actual remaining lifespan of the robot's key components or the entire robot, and is used as label data for model training. y_pred is the predicted remaining lifetime value, representing the estimated remaining lifetime value calculated by the model using feature data.

[0112] 4. If the MSE does not meet the requirements, use the grid search method to traverse the parameter combinations (C: 1.0 / 3.0 / 5.0 / 10.0; γ: 0.01 / 0.05 / 0.1 / scale) and select the parameter combination corresponding to the minimum MSE of the test set as the optimal parameter.

[0113] 5. Retrain the model with the optimal parameters, and divide 20% of the training set as the validation set. The validation set MSE should be ≤0.03 to ensure model stability.

[0114] The steps for predictive application are as follows:

[0115] 1. Perform standardization and dimensionality reduction processing on the real-time acquired performance characteristics and environmental factors, consistent with the training data.

[0116] 2. Input the preprocessed new data into the optimized SVR model and output the predicted remaining lifetime value.

[0117] 3. If the predicted value deviates from the historical degradation trend by more than 10%, the extrapolation results of the linear degradation trajectory model are combined with weighted fusion, where the SVR predicted value has a weight of 0.7 and the linear model has a weight of 0.3.

[0118] III. The specific implementation of the neural network algorithm is as follows:

[0119] A three-layer BP neural network (input layer - hidden layer - output layer) is adopted, which has a simple structure, is easy to train, and is suitable for the real-time requirements of robot life prediction.

[0120] The input layer has the same number of neurons as the feature dimension n+m, and is used to receive standardized feature data. The hidden layer has 2*(n+m) neurons, and uses the ReLU activation function, f(x)=max(0,x), which solves the vanishing gradient problem and improves training efficiency. The number of hidden layers can be adjusted to 1-2 layers depending on the feature complexity; when the feature dimension is ≥10, one additional layer is added, with n+m neurons. The output layer has 1 neuron, using a linear activation function (f(x)=x), and is used to output a continuous remaining lifetime value.

[0121] Where f(x) is the output value of the activation function, which performs a nonlinear transformation on the input of the hidden layer of the neural network, enhances the model's fitting ability, and suppresses gradient vanishing; x is the input value of the hidden layer neurons, which is the weighted sum of the neurons in the hidden layer of the neural network.

[0122] The training parameters are set as follows: the loss function uses mean squared error (MSE), consistent with the model evaluation metric, to ensure that the training objective and performance requirements are aligned; the optimizer is Adam, with adaptive learning rate adjustment, and faster convergence speed than SGD; the initial learning rate is 0.001-0.01, with a default value of 0.005. If the loss function decreases slowly in the later stages of training, it can be adjusted to 0.001; the number of iterations (Epoch) is 1000-5000, with a default value of 2000, using early stopping, stopping training when the validation set loss does not decrease for 50 consecutive Epochs; the batch size is 32-64, with a default value of 32, balancing training speed and memory usage; L2 regularization (weight decay) is used, with a regularization coefficient of 0.0001-0.001 to prevent overfitting.

[0123] The training and optimization process is as follows:

[0124] 1. Construct the above three-layer BP neural network structure and initialize the weight parameters using the Xavier normal distribution.

[0125] 2. Input the training set data, calculate the output of each layer, obtain the predicted remaining lifetime, and complete the forward propagation.

[0126] 3. Calculate the gradient based on the loss function, update the weights and biases of each layer through the Adam optimizer, and complete the backpropagation.

[0127] 4. Calculate the MSE using the test set every 100 epochs, with a target MSE ≤ 0.04.

[0128] 5. If the MSE does not meet the requirements, adjust the number of hidden layer neurons (±20%), learning rate (±50%), or regularization coefficient, and retrain until the requirements are met.

[0129] 6. Solidify the optimized model parameters (weights, biases) for real-time prediction.

[0130] The prediction application steps are the same as those of the support vector machine algorithm. The same preprocessing is required for the new data, and the prediction results can be fused and corrected according to the deviation.

[0131] IV. The specific logic for model selection and switching is as follows:

[0132] When performance degradation characteristics show an approximately linear relationship with remaining life (such as joint torque retention rate and battery range degradation), a linear degradation trajectory model + SVM optimization is preferred, and the kernel function of the SVM can be switched to a linear kernel. When the degradation law is significantly nonlinear (such as multi-component collaborative degradation under high temperature and complex terrain), a BP neural network optimization is selected, or a weighted fusion prediction of "neural network + SVM" is used, with the weights adaptively allocated based on the MSE of the test set.

[0133] V. Combined Application with Performance Degradation Comprehensive Evaluation Equation

[0134] The lifetime prediction module first calculates the performance degradation index D(t) at time t using the comprehensive performance degradation evaluation equation. Then, the performance parameters (S_i(t)) and environmental factors (F_j(t)) involved in the calculation process are processed according to the aforementioned data preprocessing specifications and input into a neural network or SVM model for training and optimization. The degradation state is quantified by the equation, and the prediction accuracy is improved by combining it with the algorithm model. When D(t) exceeds a preset threshold (e.g., 1.0), an early warning or lifetime termination judgment is triggered. If the threshold is not exceeded, the remaining lifetime is extrapolated by the optimized model, forming a dual guarantee of equation quantification and algorithm optimization.

[0135] Working principle:

[0136] The working principle of this invention is to transform the reliability testing of humanoid robots from the traditional single "pass / fail" judgment to the continuous monitoring and modeling of the robot's performance "degradation process", and to achieve comprehensive and efficient reliability assessment and life prediction through a multi-unit collaborative, data-driven closed-loop architecture.

[0137] The system is centered on a central control unit, which coordinates the collaborative work of various functional units: the test scenario construction unit integrates physical environment simulation and task type definition to build a comprehensive test scenario library that closely resembles real-world applications, providing a diverse foundation for testing; the test case generation unit, based on mathematical statistics principles, uses orthogonal experimental design to select representative test cases from multi-factor, multi-level combinations of operating conditions, ensuring coverage while reducing test redundancy; the automated test execution unit uses multi-sensor fusion modules deployed on key parts of the robot to collect multi-dimensional data from core components such as joints, motors, and power supplies in real time, and works with the motion control module to drive the robot to automatically execute tests according to test cases; the data processing and analysis unit cleans the collected raw data, identifies sudden failure modes by judging whether key parameters exceed thresholds, and establishes a performance degradation trajectory model for progressively changing parameters; the reliability assessment and life prediction unit, based on failure records and degradation models, uses reliability engineering theory to calculate quantitative indicators and predicts remaining service life through model extrapolation, ultimately forming a complete test-analysis-assessment-prediction closed loop.

[0138] How to use:

[0139] 1. Construct a comprehensive test scenario library: First, build a physical environment library covering different conditions such as temperature and humidity, ground material, slope, and obstacle distribution. Then, define a task type library including continuous walking, repeated climbing up and down stairs, weighted grabbing and placing, maintaining balance under interference, and turning in narrow passages. Combine physical environment conditions with task types to form a comprehensive test scenario library covering a variety of actual working conditions.

[0140] 2. Generate test case set: Based on the constructed comprehensive test scenario library, identify the key factors affecting the reliability of the humanoid robot and the different levels of each factor, select a suitable orthogonal array, and use orthogonal experimental design to screen representative working condition combinations to generate a high-coverage and efficient test case set.

[0141] 3. Automated execution test: The central control unit issues scheduling instructions, and the motion control module drives the humanoid robot to perform corresponding actions in sequence according to the test case set. At the same time, the multi-sensor fusion module deployed in key parts of the robot synchronously collects status data and performance data such as joint angle, torque, temperature, current and voltage, vibration, and visual processing delay in real time.

[0142] 4. Data Processing and Degradation Analysis: The collected raw data is preprocessed by cleaning, filtering, and noise reduction to extract time-domain, frequency-domain, or time-frequency-domain features. Then, it is determined whether the key parameters exceed the preset threshold. If they do, they are marked as sudden failures and the failure mode is recorded. If the parameters do not exceed the threshold but show gradual changes, a linear or exponential performance degradation trajectory model is established based on the extracted features.

[0143] 5. Reliability Assessment and Lifespan Prediction: Combining failure mode records and performance degradation trajectory models, reliability engineering theory is used to calculate quantitative indicators such as mean time between failures (MTBF) and reliability to complete the reliability assessment. At the same time, by extrapolating the performance degradation trajectory to a preset failure threshold, the remaining lifespan of the robot and key components is predicted, and a report containing assessment results and maintenance recommendations is generated.

[0144] In summary, the humanoid robot reliability testing method and system provided in this embodiment have the following advantages: By constructing a multi-unit collaborative systematic architecture and a data-driven testing closed loop, this invention specifically addresses many shortcomings of existing technologies and achieves significant technical effects.

[0145] In terms of test comprehensiveness, the integrated test scenario library incorporates diverse physical environments and task types, simulating real and complex working conditions. This effectively exposes potential defects in the robot's environmental adaptability and multi-subsystem collaborative operation, significantly improving the realism and coverage of the tests. Regarding test efficiency, the application of orthogonal experimental design significantly reduces the number of test cases. Combined with automated test execution processes, it avoids redundancy and inefficiency of manual operations, shortens the test cycle, reduces test costs, and meets the needs of rapid product iteration. In terms of failure identification and assessment, multi-sensor fusion enables comprehensive acquisition of multi-dimensional data, and data preprocessing and feature extraction are accurate and efficient. By ensuring data quality, performance degradation modeling can accurately capture sudden failures and gradual performance degradation, replacing the traditional judgment method that relies on human experience, and realizing an objective quantitative assessment of reliability. In terms of maintenance and application value, the life prediction function based on the degradation model breaks the passive situation of "post-event maintenance" or fixed-cycle maintenance, provides a scientific basis for preventive maintenance, and reduces long-term application risks. At the same time, the centralized control architecture and intelligent adaptive control strategy ensure the stable and coordinated operation of the system. The overall solution has universality and scalability, can be adapted to different types of robots, and provides complete technical support for improving robot reliability.

[0146] In addition, this embodiment also provides the following specific examples:

[0147] I. Technical Solution

[0148] (I) System Composition

[0149] The architecture of the humanoid robot reliability testing system in this embodiment specifically includes:

[0150] 1. Central Control Unit: It adopts an industrial computer as the core, equipped with a real-time operating system, and establishes communication with each unit through the CAN bus to realize unified scheduling of the test process, parameter configuration and abnormal response, and supports intelligent decision-making and adaptive control.

[0151] 2. Test Scenario Construction Unit:

[0152] The physical environment simulation module integrates a temperature and humidity chamber (for low, normal, and high temperature regulation), a replaceable ground platform (with three materials: tile, carpet, and gravel), a six-degree-of-freedom motion platform (simulating flat ground, a 15-degree slope, and stairs), and an obstacle simulation device (including fixed obstacles and speed-adjustable moving obstacles). Each environmental simulation unit achieves precise switching of temperature, humidity, ground condition, and slope through a collaborative control protocol.

[0153] Task scenario definition module: It has five preset task types: continuous walking, repeated climbing up and down stairs, 5kg weight grabbing and placing, maintaining balance under slight impact interference, and turning around in a 1.2-meter narrow passage. It supports the configuration of task parameters (such as walking distance and number of times climbing up and down stairs).

[0154] 3. Test Case Generation Unit: Based on orthogonal experimental design, three key influencing factors are selected: temperature, ground material, and task type. Each factor has three levels (temperature: low temperature, normal temperature, high temperature; ground material: tile, carpet, gravel; task type: walking on flat ground, climbing stairs, lifting and placing with weight). An L9(3^4) orthogonal array is used to generate 9 sets of high-coverage test cases. The specific test case combinations are as follows:

[0155] Use Case 1: Low temperature + tile + walking on flat ground;

[0156] Use Case 2: Low temperature + carpet + stairs;

[0157] Use Case 3: Low Temperature + Sand and Gravel + Heavy-Duty Grabbing and Placement;

[0158] Use Case 4: Room temperature + ceramic tiles + stairs;

[0159] Use Case 5: Room temperature + carpet + weighted gripping and placement;

[0160] Use Case 6: Walking on normal temperature, gravel, and flat ground;

[0161] Use Case 7: High Temperature + Tiles + Load-Bearing Gripping and Placement;

[0162] Use Case 8: High Temperature + Carpet + Walking on Flat Ground;

[0163] Use Case 9: High Temperature + Sand and Gravel + Going Up and Down Stairs;

[0164] 4. Automated Test Execution Unit:

[0165] Multi-sensor fusion module: Sensors are deployed in key parts of the robot, such as the knee joint, hip joint, drive motor, battery pack, and vision processing module. These include joint encoders (to collect joint angles), torque sensors (to collect joint torque), temperature sensors (to collect the temperature of motors and electronic components), current and voltage sensors (to collect battery charging and discharging parameters), IMU sensors (to collect vibration and robot posture), and industrial cameras (to collect vision processing delay). The data acquisition frequency is set to 100Hz.

[0166] Motion control module: Receives instructions from the central control unit and drives the robot's joint motors and walking mechanism to perform actions according to the test cases, achieving accurate reproduction of the actions.

[0167] 5. Data Processing and Analysis Unit:

[0168] Data preprocessing module: The Kalman filter algorithm is used to reduce noise in the raw data, and outliers are removed by the 3σ criterion to complete the data cleaning.

[0169] Degradation analysis module: Extracts time-domain features (mean, peak value, variance) and frequency-domain features (power spectral density) of data, sets failure thresholds (such as motor temperature exceeding the preset upper limit, joint torque fluctuation exceeding the set range), identifies sudden failure modes; and establishes a linear performance degradation trajectory model for parameters that have not failed but whose performance is gradually changing (such as joint torque retention rate, battery endurance).

[0170] 6. Reliability Assessment and Lifetime Prediction Unit:

[0171] Reliability index calculation module: Based on failure records and degradation data, calculate quantitative indicators such as mean time between failures and reliability.

[0172] Life prediction module: Extrapolates the linear degradation model to a preset failure threshold (such as the joint torque retention rate dropping to 80%) to predict the remaining life of key components and the whole machine; uses neural network algorithms to train and optimize the model to improve prediction accuracy.

[0173] (II) Test Methods and Procedures

[0174] 1. Construct a comprehensive test scenario library: Start the physical environment simulation module to complete the deployment of three types of temperature and humidity, three types of ground materials, three types of slope and obstacles; activate five preset task types through the task scenario definition module to form a comprehensive test scenario library that includes combinations of environment and task.

[0175] 2. Generate test case set: The test case generation unit reads the parameters of the comprehensive test scenario library and automatically generates 9 sets of test cases based on the L9(3^4) orthogonal array to ensure coverage of the main working condition combinations.

[0176] 3. Automated test execution: The central control unit issues test instructions in sequence, and the motion control module drives the humanoid robot to execute 9 sets of test cases in sequence; at the same time, the multi-sensor fusion module collects status data and performance data of each part in real time and transmits them to the central control unit for storage.

[0177] 4. Data Processing and Degradation Analysis: The data preprocessing module performs noise reduction and anomaly removal on the collected data; the degradation analysis module extracts feature parameters, determines whether there are sudden failures and records the patterns, and constructs a linear degradation model for the gradually changing parameters.

[0178] 5. Reliability Assessment and Life Prediction: The reliability index calculation module outputs indicators such as mean time between failures (MTBF) and reliability; the life prediction module extrapolates through a degradation model to obtain the remaining life of key components such as the knee joint reducer and battery, and generates a report containing assessment results and maintenance recommendations; the central control unit adaptively adjusts subsequent test parameters based on real-time data (e.g., if the motor temperature is too high in a certain test case, the test duration for that condition is appropriately shortened).

[0179] II. Working Principle

[0180] This example works by employing a closed-loop architecture with multi-unit collaboration to shift from "pass / fail" judgments to "performance degradation monitoring." The central control unit acts as the central coordinating unit, ensuring the orderly operation of all units. First, the test scenario construction unit builds a foundation of complex operating conditions closely resembling real-world applications, addressing the issue of limited test scenarios in traditional methods. The test case generation unit, based on orthogonal experimental design, reduces redundant testing while maintaining coverage, improving efficiency. The automated test execution unit achieves comprehensive multi-dimensional data collection through multi-sensor fusion, providing reliable data support for subsequent analysis. The data processing and analysis unit cleanses the data, accurately identifying sudden failures and progressive degradation, replacing traditional manual judgment methods. Finally, the reliability assessment and lifespan prediction unit, based on the analysis results, performs quantitative assessments and lifespan predictions, providing a basis for preventative maintenance. The entire process forms a data-driven closed loop of "scenario construction - test case generation - data collection - analysis and modeling - assessment and prediction," ensuring the systematic and scientific nature of the testing.

[0181] III. Experimental Data

[0182] During the test, all 9 test cases were executed successfully without any interruption, and the data integrity rate of the sensor acquisition reached 100%.

[0183] Failure mode identification results: Sudden failures occurred in 3 test cases, all of which were due to the motor temperature exceeding the threshold under high temperature conditions, and the corresponding failure mode was "motor overheat protection triggered".

[0184] Performance degradation trend: The joint torque retention rate gradually decreased over the test time. The degradation rate was fastest under the conditions of gravel floor + high temperature + going up and down stairs, while the degradation rate was slowest under the conditions of tile floor + normal temperature + walking on flat ground.

[0185] The primary and secondary factors affecting the wear and tear of joints were determined through range analysis. Ground material had the greatest impact on joint wear, followed by temperature, while task type had a relatively smaller impact.

[0186] Reliability indicators: The mean time between failures (MTBF) calculated based on test data shows a stable trend, and the reliability meets the preset evaluation standards; life prediction results show that the remaining life of the knee joint reducer is shorter than that of other components, which requires special attention.

[0187] IV. Technical Effects

[0188] Comprehensive testing: By constructing a comprehensive scenario library containing various temperatures, humidity levels, ground conditions, slopes, and tasks, the system simulates real and complex working conditions, effectively exposing the robot's potential defects under harsh conditions such as high temperatures and rough ground. This solves the problem of the limited range of traditional testing scenarios and improves the realism and coverage of the tests.

[0189] Testing efficiency: The orthogonal test method was used to reduce the original large number of test cases to 9 sets. Combined with automated test execution, it avoided the inefficiency and redundancy of manual operation, greatly shortened the test cycle, and met the needs of rapid product iteration development.

[0190] Failure identification accuracy: Multi-sensor fusion enables comprehensive acquisition of multi-dimensional data, data preprocessing and feature extraction ensure data quality, and performance degradation modeling can accurately capture sudden failures and gradual performance degradation, replacing the traditional judgment method that relies on human experience, and realizing the objectivity and accuracy of failure identification.

[0191] Scientific assessment and prediction: By calculating quantitative indicators such as mean time between failures (MTBF) and reliability, an objective assessment of reliability is achieved; the life prediction function based on the degradation model breaks the passive situation of "post-event maintenance", provides a scientific basis for preventive maintenance of key components, and reduces the risk of long-term robot application.

[0192] System stability: The centralized control architecture enables collaborative work among all units, and the intelligent adaptive control strategy can adjust test parameters based on real-time data, ensuring stable operation of the test process and improving the reliability and repeatability of test results.

[0193] The above are merely preferred embodiments of the present invention and are not intended to limit the implementation methods and protection scope of the present invention. Those skilled in the art should recognize that any equivalent substitutions and obvious changes made based on the description and illustrations of the present invention should be included within the protection scope of the present invention.

Claims

1. A humanoid robot reliability testing system, characterized by, It includes a central control unit and, respectively, a test scenario construction unit, a test case generation unit, an automated test execution unit, a data processing and analysis unit, and a reliability assessment and lifespan prediction unit, all of which are communicatively connected to the central control unit, wherein: The central control unit is used to schedule the collaborative work of various units to achieve unified control of the testing process; the test scenario construction unit is used to build a comprehensive test scenario library covering various actual working conditions; the test case generation unit is used to generate a high-coverage test case set based on the comprehensive test scenario library; the automated test execution unit is used to drive the humanoid robot to execute test cases and collect various data during the robot's operation in real time; the data processing and analysis unit is used to process the collected data, identify failure modes, and analyze performance degradation trends; the reliability assessment and life prediction unit is used to conduct a quantitative reliability assessment based on the analysis results and predict the remaining service life of the robot and key components. The test scenario construction unit includes a physical environment simulation module and a task scenario definition module. The physical environment simulation module is used to simulate physical environments with different temperatures and humidity, ground materials, slopes, and obstacle distributions. The task scenario definition module is used to define task types such as walking, going up and down stairs, grabbing and placing, and turning in narrow passages. The physical environment simulation module and the task scenario definition module work together to form a comprehensive test scenario library that includes combinations of environment and task. The test case generation unit is configured to use orthogonal experimental design method, taking physical environment type, task complexity and robot motion mode as factors, setting multiple levels for each factor, obtaining multiple factors affecting the reliability of humanoid robot and different levels of each factor, selecting the corresponding orthogonal table, extracting working condition combinations from the comprehensive test scenario library, and generating a test case set covering the level combinations of each factor. The reliability assessment and life prediction unit includes a reliability index calculation module and a life prediction module. The reliability index calculation module is used to calculate the mean time between failures (MTBF) and reliability index based on failure mode records and performance degradation data. The life prediction module is used to extrapolate the performance degradation trajectory to a preset failure threshold to obtain the remaining service life of the robot and key components. The life prediction module uses a comprehensive performance degradation evaluation equation to quantify the degradation state of the robot as a whole or key components. The equation is as follows: ； in: D(t) represents the performance degradation exponent at time t; This represents the measured value of the i-th key performance parameter at time t; This indicates the initial health status value of the corresponding parameter; Indicates the cumulative task execution time; Indicates the design lifespan of the task; This represents the real-time intensity of the j-th type of environmental factor; This represents the design threshold for the corresponding environmental factor; α, β, γ are weighting coefficients, obtained through training with historical data; These are the environmental factor weights, reflecting the degree of impact of different environments on degradation; n and m represent the number of performance parameters and environmental factors, respectively.

2. The humanoid robot reliability testing system according to claim 1, characterized in that, The automated test execution unit includes a multi-sensor fusion module and a motion control module. The multi-sensor fusion module is deployed at key parts of the humanoid robot, such as joints, motors, power supply, and sensing components, and is used to collect data on joint angles, torque, temperature, current and voltage, vibration, and visual processing delay. The motion control module is used to receive instructions from the central control unit and drive the humanoid robot to perform corresponding actions according to the test cases.

3. The humanoid robot reliability testing system according to claim 1, characterized in that, The data processing and analysis unit includes a data preprocessing module and a degradation analysis module; the data preprocessing module is used to clean, filter, and reduce noise in the collected raw data. The degradation analysis module is used to determine whether key parameters in the data exceed preset thresholds, identify and record sudden failure modes, and establish a performance degradation trajectory model for parameters that change gradually.

4. The humanoid robot reliability testing system according to claim 1, characterized in that, The central control unit adopts a centralized control architecture and establishes data interaction with each unit through a preset communication protocol to realize intelligent decision-making for test parameter configuration, test process scheduling, and abnormal situation response, as well as adaptive adjustment of test strategies based on the robot's real-time operating status.

5. A method for testing the reliability of a humanoid robot based on the humanoid robot reliability testing system according to any one of claims 1-4, characterized in that, Includes the following steps: Step 1: Build a comprehensive test scenario library, integrating physical environment and task scenarios to form a set of test scenarios covering various actual working conditions; Step 2: Based on the comprehensive test scenario library, use experimental design methods to generate a high-coverage test case set; Step 3: Automated drive the humanoid robot to execute the test case set, synchronously collecting various status data and performance data during the robot's operation; Step 4: Process the collected data, identify failure modes, and establish a performance degradation trajectory model; Step 5: Based on the failure modes and performance degradation trajectory model, conduct a quantitative reliability assessment and predict the remaining service life of the robot and key components.

6. The humanoid robot reliability testing method according to claim 5, characterized in that, The process of constructing a comprehensive test scenario library in Step 1 includes: constructing a physical environment library covering low temperature, normal temperature, high temperature, low humidity, normal humidity, high humidity, smooth ground, rough ground, gravel ground, flat ground, slope, and stairs; defining a task type library including continuous walking, repeated climbing up and down stairs, weighted grasping and placement, maintaining balance under disturbance, and turning around in narrow passages; and combining the environmental conditions in the physical environment library with the task types in the task type library to form a comprehensive test scenario library.

7. The humanoid robot reliability testing method according to claim 5, characterized in that, Step four involves establishing a performance degradation trajectory model, which includes: extracting features from the preprocessed data, including time-domain features, frequency-domain features, or time-frequency-domain features; determining whether key parameters exceed preset thresholds, and if so, marking them as sudden failures and recording the failure mode; if the parameters do not exceed the thresholds but show gradual changes, then establishing a linear or exponential performance degradation trajectory model based on the extracted features; and in step five, the lifetime prediction process uses neural network or support vector machine algorithms to train and optimize the performance degradation trajectory model to improve the accuracy of remaining lifetime prediction.

Images