Methods, apparatus and equipment for testing the performance of humanoid robot joint modules

Abstract

This application discloses a method, apparatus, and equipment for testing the performance of a humanoid robot joint module, relating to the field of robotics technology. The method includes: acquiring the identity identifier of the joint module under test; loading corresponding standard test procedures and digital twin models from the cloud based on the identity identifier; controlling the test platform to execute an automated test sequence based on the standard test procedures; synchronously collecting multimodal performance data of the joint module under test through multiple sensors and recording corresponding timestamps; running a lightweight AI model in real time through an edge computing unit to perform preliminary analysis of the multimodal performance data and adjust the test strategy based on the preliminary analysis results; uploading the multimodal performance data with the identity identifier and timestamps to a cloud data lake; performing in-depth analysis of the multimodal performance data through a cloud-based AI analysis engine to generate a test report. This application achieves intelligent testing by constructing an "edge-cloud" collaborative intelligent testing system.

Description

Technical Field

[0001] This application relates to the field of robotics technology, and in particular to a method, apparatus and equipment for testing the performance of a humanoid robot joint module. Background Technology

[0002] With the rapid development of humanoid robot technology, the performance and reliability of its joint modules have become crucial factors determining the overall performance of the robot. As a precision system integrating motors, reducers, drives, and controllers, the joint module exhibits complex dynamic characteristics and diverse failure modes, placing extremely high demands on the comprehensiveness and depth of testing. Ensuring its stability, accuracy, and lifespan under various operating conditions is a core technological aspect for improving robot product quality and market competitiveness.

[0003] Currently, most performance tests of joint modules in the industry follow pre-written, fixed test scripts. Regardless of the module's performance during testing, the system mechanically completes all preset procedures, resulting in a rigid testing process lacking intelligent adaptive capabilities. Furthermore, test data primarily focuses on basic physical quantities such as torque, position, speed, and temperature. Data analysis is mostly descriptive, generating performance curves and comparing them with theoretical values ​​or single thresholds, resulting in a limited data dimension. Test data is typically stored on local industrial control computers or LAN servers, leading to inconsistent data formats, difficulty in interconnecting with other systems, system isolation, and data silos.

[0004] Therefore, existing fixed testing procedures cannot be dynamically adjusted or continuously updated according to the real-time status of the robot's tested joint modules to cope with increasingly complex failure modes and ever-improving product quality requirements.

[0005] The above content is only used to help understand the technical solution of this application and does not represent an admission that the above content is prior art. Summary of the Invention

[0006] The main purpose of this application is to provide a method, apparatus and equipment for testing the performance of humanoid robot joint modules, which aims to solve the technical problem of not being able to dynamically adjust and continuously update the testing strategy and judgment criteria according to the real-time status of the robot joint module under test.

[0007] To achieve the above objectives, this application proposes a method for testing the performance of a humanoid robot joint module, the method comprising:

[0008] Obtain the identity identifier of the joint module under test, and load the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier;

[0009] Based on the standard test procedure, the control test platform executes an automated test sequence, synchronously collects multimodal performance data of the tested joint module through multiple sensors, and records the corresponding timestamps;

[0010] A lightweight AI model is run in real time through an edge computing unit to perform preliminary analysis on the multimodal performance data, and the testing strategy is adjusted based on the preliminary analysis results.

[0011] Multimodal performance data with identification and timestamps is uploaded to a cloud data lake, and the cloud-based AI analysis engine performs in-depth analysis on the multimodal performance data to generate a test report.

[0012] In one embodiment, the step of synchronously acquiring multimodal performance data of the joint module through multiple sensors includes:

[0013] The current harmonic data of the tested joint module are simultaneously measured using a high-precision electrical parameter analyzer.

[0014] The positioning error data of the tested joint module was measured using a high-resolution optical encoder.

[0015] Vibration signals of the outer shell of the tested joint module were collected using a high-frequency vibration accelerometer.

[0016] The audio signal of the tested joint module during operation is acquired by an acoustic microphone or an ultrasonic sensor.

[0017] The temperature field distribution data of the tested joint module is acquired non-contactly using a thermal imaging camera.

[0018] In one embodiment, the step of performing preliminary analysis of the multimodal performance data using a real-time lightweight AI model includes:

[0019] The synchronously acquired current harmonic data, positioning error data, vibration signal, audio signal and temperature field distribution data are time-aligned.

[0020] A lightweight AI model based on an attention mechanism is used to dynamically assign weights to the multimodal performance data and fuse them to generate a real-time health status vector.

[0021] The real-time health status vector is compared and learned with the standard health status vector space of the digital twin model, and the relative distance or likelihood probability of the real-time health status vector in the standard health status vector space is calculated.

[0022] If the relative distance exceeds a preset distance threshold or the likelihood probability is lower than a preset probability threshold, an anomaly is determined, and one or more preliminary hypothetical diagnoses are generated based on the deviation of each component in the real-time health status vector.

[0023] In one embodiment, the step of adjusting the testing strategy based on the preliminary analysis results includes:

[0024] The initial hypothesis diagnosis and the current test procedure context are input into a test policy generator pre-trained based on a meta-learning framework;

[0025] Based on the test strategy generator, verification test sequences are dynamically selected and combined from the parameterized atomic test instruction library;

[0026] The confirmatory test sequence is inserted into the current test procedure and executed to verify whether the preliminary diagnostic hypothesis is valid.

[0027] In one embodiment, the step of adjusting the testing strategy based on the preliminary analysis results includes:

[0028] If the preliminary diagnostic hypothesis is true, a high-priority alarm is generated, triggering a convergence test strategy. Based on the preliminary diagnostic hypothesis, the executed confirmatory test sequence, and the verification results, a triplet is constructed and uploaded to the cloud to optimize and update the test strategy generator.

[0029] If the initial diagnostic assumptions are not valid, then the process should revert to the standard testing procedure.

[0030] In one embodiment, the step of performing in-depth analysis of the multimodal performance data using a cloud-based AI analysis engine includes:

[0031] The uploaded multimodal performance data with identification and timestamps are spatiotemporally correlated and matched with historical test data of the same type of joint module stored in the cloud data lake to construct a spatiotemporally correlated dataset.

[0032] Feature extraction and abnormal pattern detection are performed on the spatiotemporal correlated dataset using a pre-trained multimodal fusion model in the cloud.

[0033] Based on the results of feature extraction and abnormal pattern detection, the performance degradation index and remaining service life prediction of the tested joint module are calculated.

[0034] In one embodiment, the step of performing feature extraction and abnormal pattern detection on the multimodal performance data using a pre-trained multimodal fusion model in the cloud includes:

[0035] The spatiotemporal correlated dataset is input into a hierarchical attention fusion network, and feature encoding is performed on the data of each modality separately;

[0036] The features of each modality are dynamically weighted and fused through an attention mechanism, and a unified multimodal feature representation is output by combining the feature encoding.

[0037] Anomaly detection algorithms are used to identify anomalous points in the multimodal feature representation that deviate from the normal pattern.

[0038] In one embodiment, after the step of performing feature extraction and abnormal pattern detection on the multimodal performance data using a pre-trained multimodal fusion model in the cloud, the following steps are included:

[0039] The outliers are compared with the ideal performance data simulated by the digital twin model to calculate a multi-dimensional deviation vector, and the parameters of the digital twin model are calibrated based on the multi-dimensional deviation vector.

[0040] Furthermore, to achieve the above objectives, this application also proposes a performance testing device for a humanoid robot joint module, the humanoid robot joint module performance testing device comprising:

[0041] The model loading module is used to obtain the identity identifier of the joint module under test, and load the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier;

[0042] The data acquisition module is used to control the test platform to execute automated test sequences based on the standard test procedure, synchronously acquire multimodal performance data of the tested joint module through multiple sensors, and record the corresponding timestamps;

[0043] The data processing module is used to run a lightweight AI model in real time through the edge computing unit, perform preliminary analysis on the multimodal performance data, and adjust the test strategy based on the preliminary analysis results.

[0044] The cloud processing module is used to upload multimodal performance data with identification and timestamps to the cloud data lake, and to perform in-depth analysis of the multimodal performance data through the cloud AI analysis engine to generate test reports.

[0045] In addition, to achieve the above objectives, this application also proposes a humanoid robot joint module performance testing device, the device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the humanoid robot joint module performance testing method described above.

[0046] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the humanoid robot joint module performance testing method described above.

[0047] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the humanoid robot joint module performance testing method described above.

[0048] One or more technical solutions proposed in this application have at least the following technical effects:

[0049] This application obtains the identity identifier of the tested joint module, loads the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier, controls the test platform to execute an automated test sequence based on the standard test procedure, synchronously collects multimodal performance data of the tested joint module through multiple sensors, and records the corresponding timestamps, runs a lightweight AI model in real time through an edge computing unit to perform preliminary analysis on the multimodal performance data, and adjusts the test strategy based on the preliminary analysis results, and uploads the multimodal performance data with identity identifier and timestamp to the cloud data lake, performs in-depth analysis on the multimodal performance data through the cloud AI analysis engine, and generates a test report. This achieves standardized management and unified personalized configuration of the test process, significantly improving the flexibility and applicability of the test system; the test process undergoes a fundamental shift from single-dimensional judgment to multimodal fusion diagnosis, and from post-analysis to real-time insight, greatly improving the depth and timeliness of defect detection; the adaptive and dynamic optimization of the test strategy allows test resources to focus on suspicious issues, thereby significantly improving the efficiency and targeting of the test, and ensuring that the judgment benchmark and predictive ability of the test system can continuously improve over time. Attached Figure Description

[0050] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 This is a flowchart illustrating the first embodiment of the performance testing method for humanoid robot joint modules proposed by the applicant.

[0053] Figure 2 This is a flowchart illustrating Embodiment 2 of the performance testing method for humanoid robot joint modules provided by the applicant.

[0054] Figure 3This is a flowchart illustrating Embodiment 3 of the performance testing method for humanoid robot joint modules provided by the applicant.

[0055] Figure 4 This is a flowchart illustrating Example 4 of the applicant's method for testing the performance of humanoid robot joint modules.

[0056] Figure 5 This is a schematic diagram of the module structure of the humanoid robot joint module performance testing device according to an embodiment of this application;

[0057] Figure 6 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the performance testing method of the humanoid robot joint module in this application embodiment.

[0058] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0060] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0061] Because existing fixed testing procedures cannot be dynamically adjusted according to the real-time status of the robot's tested joint modules, and cannot continuously update testing strategies and judgment standards, they are unable to cope with increasingly complex failure modes and ever-improving product quality requirements.

[0062] This application provides a solution that obtains the identity identifier of the joint module under test, loads the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier, controls the test platform to execute automated test sequences based on the standard test procedure, synchronously collects multimodal performance data of the joint module under test through multiple sensors, and records the corresponding timestamps, runs a lightweight AI model in real time through an edge computing unit to perform preliminary analysis of the multimodal performance data, and adjusts the test strategy based on the preliminary analysis results, and uploads the multimodal performance data with identity identifier and timestamp to the cloud data lake, performs in-depth analysis of the multimodal performance data through the cloud AI analysis engine, and generates a test report. Intelligent testing is achieved by constructing an intelligent testing system that coordinates "edge-cloud" and "end-end" collaboration.

[0063] Based on this, embodiments of this application provide a method for testing the performance of a humanoid robot joint module, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the performance testing method for humanoid robot joint modules proposed by the applicant.

[0064] In this embodiment, the performance testing method for the humanoid robot joint module includes steps S10 to S40:

[0065] Step S10: Obtain the identity identifier of the joint module under test, and load the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier;

[0066] It should be noted that, in this embodiment, the identity identifier refers to a unique digital identity credential bound to a specific joint module. Its form can be a QR code, barcode, or a unique ID code stored in an RFID chip, and can be obtained through a scanning device. The standard test procedure refers to a predefined set of structured, executable test instructions for a specific model of joint module. It details the test sequence, excitation signals, safety parameters, and performance criteria. The standard test procedure may include no-load speed testing, rated load efficiency testing, step response testing, etc. The digital twin model refers to a virtual mapping of this model of joint module in the cloud. It is a dynamic, data-driven model that integrates its physical characteristics, ideal behavior model, historical performance data cluster, and standard health state vector space. Through identity recognition, the testing system can automatically call the most suitable test resources for the currently tested joint module, ensuring the accuracy and relevance of the test.

[0067] Step S20: Based on the standard test procedure, the control test platform executes an automated test sequence, synchronously collects multimodal performance data of the tested joint module through multiple sensors, and records the corresponding timestamps;

[0068] It should be noted that, in this embodiment, the test platform refers to physical equipment integrating joint module fixtures, servo drivers, load simulators, and data acquisition systems. Automated test sequences refer to a series of test actions automatically executed in a logical order, as defined by standard test procedures. Multi-sensor synchronous acquisition refers to using a unified hardware clock signal to trigger multiple different types of sensors, enabling them to acquire data under the same time reference, thus ensuring time consistency between different modal data. Multimodal performance data refers to a set of data describing the operating state of the joint module from different physical dimensions, such as vibration, sound, temperature, current, torque, and position. A timestamp refers to a precise time stamp based on a unified clock for each acquired data record. By executing standardized test sequences, the consistency of test conditions and the comparability of results are ensured; the sensor synchronous acquisition strategy overcomes the limitations of traditional single-dimensional measurement, providing a rich data foundation for comprehensive diagnosis of potential faults.

[0069] Step S30: Run a lightweight AI model in real time through the edge computing unit to perform preliminary analysis on the multimodal performance data, and adjust the test strategy based on the preliminary analysis results;

[0070] It should be noted that, in this embodiment, the edge computing unit refers to an embedded computer or industrial control computer with a certain computing power deployed at the test site for localized real-time data processing. The lightweight AI model refers to a small neural network model that has undergone optimization techniques such as model pruning, quantization, or knowledge distillation. Its characteristics include low computational cost and fast inference speed, making it suitable for deployment on resource-constrained edge devices. Preliminary analysis refers to online and rapid cleaning, feature extraction, and state assessment of the collected multimodal performance data. The test strategy refers to the decision-making logic guiding the test platform on how to execute tests, including the order of test steps and the parameters of the stimulus signals. By performing real-time analysis at the data source, the latency of data transmission to the cloud is greatly reduced, enabling the system to respond to abnormal states at the millisecond level. This endows the test system with real-time intelligent decision-making capabilities, allowing for more efficient reproduction and localization of intermittent or hidden defects.

[0071] In one possible implementation, the test strategy is adjusted by a test strategy generator pre-trained based on a meta-learning framework, which can dynamically combine atomic test instructions according to the current context.

[0072] Step S40: Upload the multimodal performance data with identification and timestamp to the cloud data lake, and perform in-depth analysis of the multimodal performance data through the cloud AI analysis engine to generate a test report.

[0073] It should be noted that, in this embodiment, the cloud data lake refers to a centralized repository in cloud storage capable of holding massive amounts of raw data (including structured and unstructured data), allowing access and processing of the data using various computing and analysis frameworks. The cloud-based AI analysis engine refers to a large-scale artificial intelligence analysis system deployed on a cloud server, possessing powerful computing capabilities, capable of performing complex batch processing, deep learning training, and large-scale data mining. Deep analysis refers to a more complex and time-consuming data analysis process compared to the preliminary analysis of edge computing units, including but not limited to batch reprocessing, long-term trend prediction, cross-device correlation analysis, and high-dimensional feature mining. The test report refers to a structured document automatically generated by the system that integrates all findings from this test, including performance indicators, health status assessments, anomaly diagnostic conclusions, and improvement suggestions. Archiving all raw data to the data lake preserves complete information for historical review and in-depth mining. Utilizing the near-limitless computing power of the cloud for deep analysis can uncover hidden patterns and long-term degradation trends that are difficult to detect at the edge.

[0074] In one possible implementation, updating the digital twin model may include: optimizing the boundary of the standard health state vector space using new normal samples; correcting the parameters of the performance degradation prediction model using confirmed fault data; and integrating validated fault diagnosis paths into a fault knowledge graph in the cloud.

[0075] This application's embodiments achieve personalized, on-demand loading of test resources through identity identification. By simultaneously acquiring data from multiple sensors and performing real-time intelligent edge analysis, the testing process has been transformed from single-dimensional judgment to multimodal deep perception, and from post-analysis to real-time interactive diagnosis, significantly improving defect detection efficiency and accuracy. Finally, by using a cloud data lake and AI engine to deeply mine all data and reverse-update the digital twin model, the dynamic improvement and long-term reliability of testing standards are ensured, constructing a closed-loop testing system that integrates "end-edge-cloud" collaboration.

[0076] In one possible implementation, the step of synchronously acquiring multimodal performance data of the joint module via multiple sensors includes:

[0077] The current harmonic data of the tested joint module are simultaneously measured using a high-precision electrical parameter analyzer.

[0078] The positioning error data of the tested joint module was measured using a high-resolution optical encoder.

[0079] Vibration signals of the outer shell of the tested joint module were collected using a high-frequency vibration accelerometer.

[0080] The audio signal of the tested joint module during operation is acquired by an acoustic microphone or an ultrasonic sensor.

[0081] The temperature field distribution data of the tested joint module is acquired non-contactly using a thermal imaging camera.

[0082] It should be noted that the process of synchronously acquiring multimodal performance data of the joint module through multiple sensors is a comprehensive data acquisition scheme integrating various high-precision sensing technologies. It aims to collect data from multiple physical domains of the tested joint module, including its electrical characteristics, accuracy characteristics, dynamic response characteristics, acoustic characteristics, and thermal characteristics. Among these, current harmonic data refers to the distortion component in the motor drive current that deviates from the standard sine wave; its spectral characteristics can effectively reflect abnormal electromagnetic characteristics within the motor, faults in the driver switching devices, or sudden load changes. Positioning error data refers to the deviation between the actual output position and the commanded position of the joint module, measured by a high-resolution optical encoder, and is a key indicator for evaluating servo control accuracy and transmission mechanism hysteresis. A high-frequency vibration accelerometer is a sensor capable of measuring high-frequency dynamic acceleration, used to capture the weak vibration characteristics generated by the internal mechanical components of the joint module under high-speed operation. The vibration signal contains rich fault characteristic information, such as the defect characteristic frequency of the bearing, the meshing frequency of the gear, and its sidebands. By installing vibration accelerometers at specific locations on the outer casing (such as the bearing housing or reducer mounting surface), the testing system can non-invasively monitor the health of the internal mechanical transmission chain, achieving early warning and precise location of potential mechanical faults. Specifically, a triaxial vibration accelerometer is used to simultaneously capture vibrational energy in three orthogonal directions to describe the vibrational characteristics.

[0083] An acoustic microphone is a sensor capable of capturing sound fluctuations in the audible frequency range (e.g., 20Hz - 20kHz) and converting them into electrical signals. An ultrasonic sensor is a sensor capable of capturing sound wave signals with frequencies higher than 20kHz, and is particularly sensitive to weak impact sound waves generated by gas leaks, micro-discharges, and early surface wear. An audio signal refers to the time-series electrical signal output by the aforementioned sensors, containing sound pressure level and frequency information. The acquisition and analysis of audio signals effectively complement vibration signals, jointly constructing the ability to perceive the state of the tested joint module system. For example, the test platform deployed an acoustic microphone array approximately 30 cm from the joint module and installed a directional ultrasonic sensor near the module's motor junction box for full-load durability testing of the tested joint module.

[0084] A thermal imaging camera is a device that generates an image of the temperature distribution of an object's surface (i.e., a thermal image) by detecting the infrared radiation emitted from the object's surface. Temperature field distribution data refers to a two-dimensional dataset presented as an image matrix, containing the temperature values ​​of each pixel on the surface of the tested joint module. It assesses the energy dissipation and thermal management status of the joint module from a thermodynamic perspective. Localized overheating is a typical manifestation of various faults such as motor insulation aging, bearing lubrication failure, and excessive gear wear. By acquiring the temperature field distribution of the entire joint module surface, the testing system can accurately locate overheating points, evaluate the effectiveness of the heat dissipation system, and monitor the dynamic changes in temperature rise. For example, an infrared thermal imager is fixedly mounted above the test platform, capturing images of the entire operating joint module from a top-down angle. During a 30-minute full-load test, the thermal imager records temperature field data at a frequency of one frame per minute.

[0085] Optionally, data acquisition from all sensors is triggered by a unified time base signal to ensure precise alignment of data points for different physical quantities. For example, the system employs a synchronization mechanism based on the IEEE 1588 Precision Clock Protocol (PTP) to provide microsecond-level time synchronization accuracy for sensors distributed in different physical locations.

[0086] Furthermore, referring to Figure 2 The second embodiment of the performance testing method for humanoid robot joint modules provided by this applicant provides a flowchart, based on the above. Figure 2 The embodiment shown further refines the step of "running a lightweight AI model in real time to perform preliminary analysis of the multimodal performance data" in step S30, including steps A201 to A204:

[0087] Step A201: Time-align the synchronously acquired current harmonic data, positioning error data, vibration signal, audio signal and temperature field distribution data;

[0088] It should be noted that vibration signal refers to the time-series voltage signal output by a high-frequency vibration accelerometer, reflecting the characteristics of mechanical vibration. Audio signal refers to the time-series voltage signal output by an acoustic microphone or ultrasonic sensor, reflecting the characteristics of acoustic waves. Temperature field distribution data refers to the two-dimensional temperature matrix sequence output by a thermal imaging camera, in frames.

[0089] Step A202: Use a lightweight AI model based on an attention mechanism to dynamically assign weights to the multimodal performance data and fuse them to generate a real-time health status vector.

[0090] It should be noted that the real-time health status vector refers to a numerical vector output after fusion, which is a digital and quantitative identifier of the current comprehensive health status of the tested joint module. By dynamically allocating weights through an attention mechanism, the model focuses on the modalities and features most relevant to the current state, suppressing interference from noise and redundant information.

[0091] Step A203: Compare and learn the real-time health status vector with the standard health status vector space of the digital twin model, and calculate the relative distance or likelihood probability of the real-time health status vector in the standard health status vector space.

[0092] It should be noted that, in this embodiment, the standard health state vector space of the digital twin model refers to a data distribution region or set defined in the digital twin model, consisting of health state vectors generated by a large number of normal joint modules of the same model under various working conditions. This represents the behavioral benchmark of the module in a normal state. Comparison learning refers to the computational process of matching and similarity assessment between the currently observed real-time health state vector and the aforementioned benchmark space. Relative distance refers to a mathematical scalar used to measure the difference between two vectors or between a vector and a set of vectors, such as Euclidean distance, Mahalanobis distance, or cosine distance. Relative distance directly reflects the degree to which the current state deviates from the normal benchmark. Likelihood probability refers to the probability that the current real-time health state vector belongs to the "normal" state set. For example, the likelihood probability can be obtained by calculating the probability density function on the standard health state vector space fitted by a Gaussian mixture model.

[0093] In one specific implementation, the edge computing unit obtains a standard health status vector space constructed from 10,000 normal samples from a digital twin model loaded in the cloud. This space is modeled using a multivariate Gaussian distribution. The system calculates the Mahalanobis distance from the currently generated real-time health status vector to the center of this multivariate Gaussian distribution, which is 5.8. Simultaneously, it calculates the likelihood probability of this vector under this distribution, which is 0.02.

[0094] Step A204: If the relative distance exceeds a preset distance threshold or the likelihood probability is lower than a preset probability threshold, an anomaly is determined, and one or more preliminary hypothetical diagnoses are generated based on the deviation of each component in the real-time health status vector.

[0095] It should be noted that, in this embodiment, the preset distance threshold and preset probability threshold refer to threshold values ​​set in advance through statistical learning or experience, used to binaryly determine whether the state of the tested joint module is abnormal. The deviation of each component in the real-time health status vector refers to analyzing the differences between the numerical dimensions of this vector and the baseline values ​​of the corresponding dimensions in the standard health status vector space. Generating a preliminary hypothesis diagnosis refers to inferring a list of potential fault causes that may lead to this abnormal mode based on the deviation data, combined with a lightweight AI model or pre-stored fault knowledge.

[0096] In one possible implementation, the system constructs a fault mode-feature component mapping table, and associates significantly deviating feature components with possible fault modes by querying this table. For example, the system presets a Mahalanobis distance threshold of 4.0 and a likelihood probability threshold of 0.05; the currently calculated distance (5.8) is greater than the threshold, and the probability (0.02) is lower than the threshold, so the state is judged to be abnormal; further analysis reveals that in the 128-dimensional health state vector, the values ​​of the 35th dimension (related to high-frequency vibration energy) and the 89th dimension (related to ultrasonic energy in a specific frequency band) are significantly higher. Based on the built-in fault-feature mapping relationship, the system generates two preliminary hypothetical diagnoses: "Hypothesis 1: Early pitting exists in the inner ring of the bearing (confidence level 65%)"; "Hypothesis 2: Slight scratches exist on the gear surface (confidence level 45%)".

[0097] This application's embodiments obtain multimodal data through time alignment, and then use an advanced attention mechanism to achieve information fusion, generating a digital vector that can accurately characterize the system's health status. By comparing the real-time health status vector with the benchmark value in the digital twin model, abnormal data is obtained, and preliminary hypothesis diagnosis is generated based on the abnormal data. This realizes automation and intelligence from multi-source heterogeneous data acquisition to real-time status assessment and preliminary root cause inference, greatly improving the real-time diagnostic capability and decision-making level of the testing system on site.

[0098] In one possible implementation, the step of adjusting the testing strategy based on the preliminary analysis results includes:

[0099] The initial hypothesis diagnosis and the current test procedure context are input into a test policy generator pre-trained based on a meta-learning framework;

[0100] Based on the test strategy generator, verification test sequences are dynamically selected and combined from the parameterized atomic test instruction library;

[0101] The confirmatory test sequence is inserted into the current test procedure and executed to verify whether the preliminary diagnostic hypothesis is valid.

[0102] It's important to note that the current test procedure context refers to the progress of the standard test procedure currently being executed by the test system, the completed test items and their results, and dynamic environmental information such as current test conditions (e.g., speed, load). The test strategy generator, pre-trained based on a meta-learning framework, is a core AI decision-making module. This module, through meta-learning training on a large number of different fault scenarios and test tasks, has gained the ability to quickly adapt to new scenarios and generate new strategies. Essentially, it learns "how to quickly generate solutions based on new problems," providing an intelligent decision-making core for the generation of dynamic test strategies. By inputting specific diagnostic hypotheses along with the real-time test environment, the test strategy generator can fully understand "what is happening" and "what is being done," thus providing a comprehensive information foundation for making accurate decisions. The meta-learning framework ensures that the generator does not simply match preset rules, but can, like an experienced engineer, quickly derive effective verification solutions when faced with new and unseen combinations of fault modes, utilizing its learned general strategy generation principles. Specifically, the test strategy generator employs a memory-based meta-learning model, generating new strategies by analogy by accessing multiple stored fault test cases.

[0103] Secondly, a parameterized atomic test instruction library refers to a predefined collection containing a large number of basic test operation units. Each atomic instruction is a minimal, independently executable test action, and its key parameters (such as target value, duration, sampling rate, etc.) are configurable. Dynamic selection and combination refer to the test strategy generator selecting relevant instructions from the atomic instruction library in real time based on the specific input problem. By instantiating, sorting, and logically connecting the parameters of these instructions, a complete and targeted test process is assembled. This transforms intelligent decision-making into an executable and customized test plan, eliminating the need for the system to pre-store all possible test processes, greatly enhancing the adaptability and scalability of the test system. Specifically, the combination process includes assigning specific parameter values ​​to the selected atomic instructions and organizing them using control structures such as sequence, branching, and looping.

[0104] A confirmatory test sequence refers to a set of temporary test instructions specifically designed to verify a particular preliminary diagnostic hypothesis, output by the test strategy generator. Inserting it into the current test procedure means that the test system's execution engine embeds this confirmatory sequence as a high-priority subtask into the existing execution flow without completely aborting the current standard test process. Immediately executing the generated confirmatory tests significantly accelerates the fault diagnosis process, and by designing specific test conditions, the signal-to-noise ratio of suspected fault characteristics can be greatly improved, thus making the diagnostic conclusions more reliable. For example, the test system's execution engine uses a preemptive task scheduling strategy, assigning the confirmatory test sequence the highest execution priority. After the confirmatory test sequence is completed, the system automatically compares the collected data with the fault characteristic database and automatically gives a conclusion of "hypothesis valid" or "hypothesis invalid" based on preset judgment rules.

[0105] In one specific implementation, if the preliminary diagnosis assumes an abnormal current harmonic, a current ripple depth test at a specific carrier frequency is automatically inserted into the test sequence; if the preliminary diagnosis assumes a structural resonance at a specific frequency, a high-resolution sinusoidal frequency sweep test near the resonance frequency is automatically triggered to accurately characterize the modal shape; if the preliminary diagnosis assumes a positioning accuracy drift, a positioning accuracy repeatability test is automatically added over the entire operating temperature range.

[0106] This application embodiment introduces an intelligent test strategy generator to transform preliminary diagnostic hypotheses into targeted verification actions, constructing a real-time closed loop of "perception-decision-action-verification". This significantly improves the diagnostic accuracy and efficiency of the test system. The system no longer mechanically traverses fixed processes, greatly accelerating the fault location process and enhancing the ability to detect hidden and intermittent defects, making performance testing truly possess the advanced features of interactive intelligent diagnosis.

[0107] In one possible implementation, the step of adjusting the testing strategy based on the preliminary analysis results is followed by:

[0108] If the preliminary diagnostic hypothesis is true, a high-priority alarm is generated, triggering a convergence test strategy. Based on the preliminary diagnostic hypothesis, the executed confirmatory test sequence, and the verification results, a triplet is constructed and uploaded to the cloud to optimize and update the test strategy generator.

[0109] If the initial diagnostic assumptions are not valid, then the process should revert to the standard testing procedure.

[0110] It should be noted that, in the embodiments of this application, the convergence testing strategy refers to a focused and efficient testing scheme adopted by the testing system after confirming the existence of a specific fault. Its goal is no longer broad probing, but rather limited but in-depth additional testing around the confirmed fault to quickly assess the severity and scope of the fault or to conduct destructive limit testing. A triple refers to a complete case record consisting of "preliminary diagnostic hypothesis, confirmatory test sequence, and verification results." Successful diagnostic cases are constructed as triples and uploaded, providing high-quality reinforcement learning samples for the cloud-based test strategy generator. For example, the convergence testing strategy includes performing short-term overload tests to assess the safety margin of the faulty component, or performing resonance tests at specific frequencies to accelerate fault development and observe its evolution.

[0111] In one specific implementation, after confirming the hypothesis of "pitting corrosion in the bearing inner ring," the system immediately triggers a flashing red light and a buzzer alarm in the control room, and displays detailed information about the fault on the monitoring screen. Simultaneously, the testing system automatically triggers a convergence test strategy: linearly increasing the rotational speed to 120% of the rated value within 5 minutes and continuously monitoring the growth trend of vibration indicators to assess the remaining life of the bearing. Finally, the system packages this successful diagnostic case (hypothesis: pitting corrosion in the bearing inner ring; verification sequence: 3600 RPM no-load operation with high-frequency sampling; verification result: fault characteristic frequencies clearly appear in the envelope spectrum, confirming the hypothesis) into a triplet data packet, and uploads it to the cloud analysis platform via a secure network connection for subsequent incremental training of the test strategy generator.

[0112] Before regressing to the standard test procedure, the system records the "hypothesis not being true" event, which can be archived as a pattern of environmental noise or normal fluctuation for subsequent data analysis. The test system resumes execution from the precise breakpoint of the inserted confirmatory sequence, ensuring the continuity of the standard test and the consistency of the data.

[0113] Furthermore, referring to Figure 3 The third embodiment of the performance testing method for humanoid robot joint modules provided by this applicant provides a flowchart, based on the above. Figure 3 The embodiment shown further refines the step of "performing in-depth analysis of the multimodal performance data through a cloud-based AI analysis engine" in step S40, including steps A301 to A303:

[0114] Step A301: Spatiotemporally correlate and match the uploaded multimodal performance data with identification and timestamps with the historical test data of the same type of joint module stored in the cloud data lake to construct a spatiotemporally correlated dataset;

[0115] It should be noted that, in the embodiments of this application, cloud data lake refers to a centralized repository that can store a large amount of structured and unstructured data, capable of accommodating massive historical performance data from numerous joint module tests; spatiotemporal correlation and matching refers to the process of comparing the test data of the currently tested module with its own historical test sequence in the time dimension, and comparing it with the test data group of other modules of the same model in the spatial dimension (i.e., module model, batch).

[0116] Specifically, the spatiotemporal correlation matching process first filters out all test records of the same model module from the data lake based on the identity identifier, then aligns the data sequences under the same test procedure according to the timestamp, and finally merges the current data with the filtered historical data according to the test stage to form an enhanced dataset for in-depth analysis.

[0117] Step A302: Feature extraction and abnormal pattern detection are performed on the spatiotemporal correlation dataset using a pre-trained multimodal fusion model in the cloud.

[0118] It should be noted that, in this embodiment, the multimodal fusion model includes at least a convolutional neural network (CNN) for processing vibration signals and temperature field distribution data, and a recurrent neural network (RNN) for processing current harmonic data and positioning error data time series. It is a deep learning network architecture capable of simultaneously processing and integrating data from different physical sensors (such as current, vibration, and temperature). This embodiment achieves intelligent and refined fault diagnosis and early warning through the multimodal fusion model.

[0119] Optionally, the multimodal fusion model adopts a hierarchical attention mechanism. First, it configures an independent feature encoding subnetwork for each type of data (such as current time series, vibration spectrum, and temperature image). Then, it dynamically assigns weights to features of different modalities and different time points through an attention fusion layer. Finally, it outputs a comprehensive anomaly probability through an anomaly scoring network.

[0120] Step A303: Based on the results of feature extraction and abnormal pattern detection, calculate the performance degradation index and remaining service life prediction of the tested joint module.

[0121] It should be noted that the performance degradation index is a normalized, comprehensive scalar indicator used to quantify the degree of performance degradation of a joint module relative to its brand-new manufacturing state. Remaining service life prediction refers to the expected remaining runtime or number of cycles of the joint module before failure, inferred by a prediction model based on its current performance state and historical degradation trajectory. In this embodiment, the feature vector representing the module's health status obtained in the feature extraction stage, and the anomaly level output from the anomaly detection stage, are input into a pre-trained regression or prediction model. This model learns the mapping relationship from health status features to performance degradation and eventual failure, obtaining the performance degradation index and remaining service life prediction results.

[0122] Optionally, the performance degradation index is calculated by combining the magnitude and rate of change of the multi-dimensional deviation vector, while the remaining useful life prediction adopts a sequence prediction model based on a long short-term memory network (LSTM). The sequence prediction model takes the historical performance sequence in the spatiotemporally associated dataset as input and predicts the future performance trajectory and the time when the failure threshold is reached.

[0123] In one possible implementation, after receiving the test data uploaded by a joint module of a specific model and serial number, the cloud-based AI analysis engine first executes step A301: Based on its identification, the system retrieves the current, vibration, and temperature data and corresponding timestamps from all 100 tests conducted on that model of module within the past year from the cloud data lake; using a time alignment algorithm, it matches the data from this test under the same operating condition of "uniform operation at rated load" with data fragments under the same operating condition in historical data, constructing a spatiotemporally correlated dataset containing the data from this test and historical group data. Next, step A302 is executed: This dataset is input into a pre-deployed multimodal fusion model. The convolutional neural network (CNN) branch in the model processes the vibration spectrum and temperature distribution images, while the recurrent neural network (RNN) branch processes the current time series. After fusion through an attention mechanism, the model outputs a feature vector and determines that there is an abnormal pattern in the current data that is highly correlated with "early bearing wear." Finally, step A303 is executed: Based on the feature vector of this abnormal mode and its position in the feature space of historical failure cases, the system calculates the current performance degradation index of the module as 0.75 (range 0-1, 1 represents brand new), and predicts its remaining service life under high-intensity workloads to be approximately 280 hours. These results are ultimately integrated into the test report.

[0124] Furthermore, referring to Figure 4 The fourth embodiment of the performance testing method for humanoid robot joint modules provided by this applicant provides a flowchart, based on the above. Figure 4The illustrated embodiment further refines the step of "extracting features and detecting abnormal patterns in the spatiotemporal correlated dataset using a pre-trained multimodal fusion model in the cloud," including steps A401-A403:

[0125] Step A401: Input the spatiotemporal correlation dataset into the hierarchical attention fusion network, and encode the features of each modality separately;

[0126] It should be noted that hierarchical attention fusion networks are a type of neural network with a hierarchical structure, used to process and understand complex data from different sources layer by layer and step by step. By extracting state information from the data of each modality, it achieves preliminary dimensionality reduction and refinement of multimodal data, transforming messy data into a set of well-organized, high-quality feature vectors.

[0127] Step A402: Dynamically weight and fuse the features of each modality through an attention mechanism, and output a unified multimodal feature representation based on the feature encoding;

[0128] It should be noted that attention mechanisms refer to techniques that dynamically and selectively focus on different parts of the input information according to the needs of the current task. Unified multimodal feature representation refers to a single or composite feature vector obtained after fusion, which integrates information from all input modalities.

[0129] Optionally, the attention mechanism calculates an attention score for each modality feature vector, which is based on the relevance of the feature vector to a learnable context vector. The score is then normalized to weights using the Softmax function. Finally, all weights are summed with the weighted sum of the corresponding feature vectors to generate a unified multimodal feature representation.

[0130] Step A403: Identify the abnormal points where the multimodal feature representation deviates from the normal pattern using an anomaly detection algorithm.

[0131] It should be noted that, in the embodiments of this application, the anomaly detection algorithm may employ a one-class support vector machine (SVM) or an isolation forest algorithm. Anomaly values ​​refer to data points in the feature space whose locations deviate significantly from the "normal pattern" region composed of a large number of normal and healthy samples, indicating that the actual operating state of the tested joint module has deviated from expectations.

[0132] Optionally, the anomaly detection algorithm used is a one-class support vector machine (SVM). This algorithm detects anomalies by finding a minimum hypersphere or hyperplane in the feature space that can contain the vast majority of normal data points. Points falling outside this region are identified as anomalies.

[0133] In one possible implementation, after the step of performing feature extraction and abnormal pattern detection on the multimodal performance data using a pre-trained multimodal fusion model in the cloud, the following steps are included:

[0134] The outliers are compared with the ideal performance data simulated by the digital twin model to calculate a multi-dimensional deviation vector, and the parameters of the digital twin model are calibrated based on the multi-dimensional deviation vector.

[0135] It should be noted that in this embodiment, the process of comparing and calibrating the detected anomalies with the ideal performance data simulated by the digital twin model constitutes a complete "monitoring-diagnosis-optimization" closed loop. The digital twin model refers to a high-fidelity dynamic simulation model of the joint module in virtual space, capable of simulating its performance output under ideal working conditions, i.e., "ideal performance data." The multi-dimensional deviation vector is a quantitative mathematical expression, where each dimension represents the difference between the actual performance data and the ideal simulation data at a specific performance index (such as current harmonic distortion rate, vibration dominant frequency amplitude, maximum temperature, etc.). This embodiment uses anomalies detected in the real world as feedback signals to evaluate and correct the accuracy of the virtual model; by calculating the multi-dimensional deviation vector, the system can quantify the severity of the current performance degradation and accurately locate which specific parameters (e.g., motor resistance, friction coefficient, transmission efficiency) drift caused the difference between the model prediction and reality. Based on these deviation vectors, the system can use parameter identification algorithms (such as least squares method, Kalman filtering) to perform reverse calibration of key physical parameters in the digital twin model.

[0136] Furthermore, this calibration mechanism transforms the digital twin model from a static, idealized reference benchmark into an adaptive model that dynamically evolves as the performance of the physical object degrades. On one hand, the calibrated model can more accurately simulate and predict the behavior of a specific joint module in its current state, providing a more reliable basis for subsequent performance prediction and health management. On the other hand, the long-term accumulated calibration data itself constitutes a valuable knowledge base, which can be used to analyze common defects and degradation patterns of this product model, thereby feeding back into the design and production stages to achieve continuous product improvement.

[0137] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the applicant's humanoid robot joint module performance testing method. Any simple modifications based on this technical concept are within the protection scope of this application.

[0138] This application also provides a performance testing device for humanoid robot joint modules. Please refer to [reference needed]. Figure 5 The humanoid robot joint module performance testing device includes:

[0139] The model loading module 10 is used to obtain the identity identifier of the joint module under test, and load the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier;

[0140] The data acquisition module 20 is used to control the test platform to execute an automated test sequence based on the standard test procedure, synchronously acquire multimodal performance data of the tested joint module through multiple sensors, and record the corresponding timestamps;

[0141] The data processing module 30 is used to run a lightweight AI model in real time through the edge computing unit, perform preliminary analysis on the multimodal performance data, and adjust the test strategy based on the preliminary analysis results.

[0142] The cloud processing module 40 is used to upload multimodal performance data with identification and timestamps to the cloud data lake, and to perform in-depth analysis of the multimodal performance data through the cloud AI analysis engine to generate a test report.

[0143] This application provides a performance testing device for humanoid robot joint modules. Employing the performance testing method for humanoid robot joint modules described in the above embodiments, it solves the technical problem of being unable to dynamically adjust and continuously update testing strategies and judgment criteria based on the real-time state of the tested robot joint modules. Compared with the prior art, the beneficial effects of the humanoid robot joint module performance testing device provided in this application are the same as those of the humanoid robot joint module performance testing method provided in the above embodiments, and other technical features in the humanoid robot joint module performance testing device are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0144] This application provides a humanoid robot joint module performance testing device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the humanoid robot joint module performance testing method in the above embodiment 1.

[0145] The following is for reference. Figure 6This document illustrates a structural schematic diagram of a humanoid robot joint module performance testing device suitable for implementing embodiments of this application. The humanoid robot joint module performance testing device in this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 6 The humanoid robot joint module performance testing equipment shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0146] like Figure 6 As shown, the humanoid robot joint module performance testing equipment may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. The random access memory 1004 also stores various programs and data required for the operation of the humanoid robot joint module performance testing equipment. The processing unit 1001, the read-only memory 1002, and the random access memory 1004 are interconnected via a bus 1005. An input / output interface 1006 is also connected to the bus. Typically, the following systems can be connected to the input / output interface 1006: input devices 1007 including, for example, a touch screen, touchpad, keyboard, mouse, image sensor, microphone, accelerometer, gyroscope, etc.; output devices 1008 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc.; storage devices 1003 including, for example, magnetic tape, hard disk, etc.; and communication devices 1009. The communication device 1009 allows the humanoid robot joint module performance testing equipment to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a humanoid robot joint module performance testing equipment with various systems, it should be understood that implementation or possession of all the systems shown is not required. More or fewer systems may be implemented alternatively.

[0147] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0148] The humanoid robot joint module performance testing equipment provided in this application, employing the humanoid robot joint module performance testing method described in the above embodiments, can solve the technical problem of being unable to dynamically adjust and continuously update the testing strategy and judgment criteria based on the real-time state of the robot joint module under test. Compared with the prior art, the beneficial effects of the humanoid robot joint module performance testing equipment provided in this application are the same as those of the humanoid robot joint module performance testing method described in the above embodiments, and other technical features in this humanoid robot joint module performance testing equipment are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0149] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0150] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. 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 scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0151] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the humanoid robot joint module performance testing method in the above embodiments.

[0152] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0153] The aforementioned computer-readable storage medium may be included in the humanoid robot joint module performance testing equipment; or it may exist independently and not assembled into the humanoid robot joint module performance testing equipment.

[0154] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by the humanoid robot joint module performance testing equipment, the humanoid robot joint module performance testing equipment: acquires the identity identifier of the joint module under test; loads the corresponding standard test procedure and digital twin model from the cloud based on the identity identifier; controls the test platform to execute an automated test sequence based on the standard test procedure; synchronously collects multimodal performance data of the joint module under test through multiple sensors and records the corresponding timestamps; runs a lightweight AI model in real time through an edge computing unit to perform preliminary analysis of the multimodal performance data and adjusts the test strategy based on the preliminary analysis results; uploads the multimodal performance data with the identity identifier and timestamps to the cloud data lake; performs in-depth analysis of the multimodal performance data through the cloud AI analysis engine; and generates a test report.

[0155] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a Local Area Network (LAN) or a Wide Area Network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0156] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0157] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0158] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described humanoid robot joint module performance testing method. This solves the technical problem of being unable to dynamically adjust and continuously update the testing strategy and judgment criteria based on the real-time state of the robot's joint module under test. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the humanoid robot joint module performance testing method provided in the above embodiments, and will not be repeated here.

[0159] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the humanoid robot joint module performance testing method described above.

[0160] The computer program product provided in this application can solve the technical problem of not being able to dynamically adjust and continuously update the testing strategy and judgment criteria based on the real-time state of the robot joint module under test. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the humanoid robot joint module performance testing method provided in the above embodiments, and will not be repeated here.

[0161] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A humanoid robot joint module performance test method, characterized by, The humanoid robot joint module performance test method comprises: Obtaining the identity of the joint module to be tested, and loading the corresponding standard test procedure and digital twin model from the cloud based on the identity; Based on the standard test procedure, control the test platform to execute the automatic test sequence, synchronously collect the multi-modal performance data of the joint module to be tested through multiple sensors, and record the corresponding time stamp, specifically including: Synchronously measure the current harmonic data of the joint module to be tested through a high-precision electrical parameter analyzer; measure the positioning error data of the joint module to be tested through a high-resolution optical encoder; collect the vibration signal of the shell of the joint module to be tested through a high-frequency vibration accelerometer; collect the audio signal of the joint module to be tested during operation through an acoustic microphone or an ultrasonic sensor; non-contact collect the temperature field distribution data of the joint module to be tested through a thermal imaging camera; Through the edge computing unit, a lightweight AI model is run in real time to preliminarily analyze the multi-modal performance data, and the test strategy is adjusted according to the preliminary analysis result, specifically including: Time align the synchronously collected current harmonic data, positioning error data, vibration signal, audio signal and temperature field distribution data; use a lightweight AI model based on an attention mechanism to dynamically allocate weights to the multi-modal performance data, fuse to generate a real-time health status vector; compare and learn the real-time health status vector with the standard health status vector space of the digital twin model, calculate the relative distance or likelihood probability of the real-time health status vector in the standard health status vector space; if the relative distance exceeds the preset distance threshold or the likelihood probability is lower than the preset probability threshold, it is determined that there is an abnormality, and one or more preliminary hypothesis diagnoses are generated based on the deviation of each component in the real-time health status vector; Input the preliminary hypothesis diagnosis and the current test procedure context to the test strategy generator based on the meta-learning framework pre-training; dynamically select and combine the verification test sequence from the parameterized atomic test instruction library based on the test strategy generator; insert the verification test sequence into the current test procedure and execute it to verify whether the preliminary hypothesis diagnosis is correct; If the preliminary hypothesis diagnosis is correct, a high-priority alarm is generated to trigger the convergence test strategy, and a triple is constructed based on the preliminary hypothesis diagnosis, the executed verification test sequence and the verification result and uploaded to the cloud to optimize and update the test strategy generator; If the preliminary hypothesis diagnosis is not correct, return to the standard test procedure for continuous execution; Upload the multi-modal performance data with identity and time stamp to the cloud data lake, and perform deep analysis on the multi-modal performance data through the AI analysis engine of the cloud to generate a test report.

2. The humanoid robot joint module performance test method according to claim 1, wherein, The step of deep analysis of the multi-modal performance data through the AI analysis engine of the cloud comprises: Spatiotemporal correlation and matching of the uploaded multi-modal performance data with identity and time stamp with the historical test data of the same type joint module stored in the cloud data lake to construct a spatiotemporal correlation dataset; characteristics extraction and anomaly pattern detection on the spatio-temporal correlated dataset through a cloud-pretrained multi-modal fusion model; based on the results of the feature extraction and anomaly pattern detection, calculate the performance degradation index and remaining useful life prediction of the joint module under test.

3. The humanoid robot joint module performance test method according to claim 2, wherein The step of performing feature extraction and anomaly pattern detection on the multi-modal performance data through a cloud-pretrained multi-modal fusion model includes: input the spatio-temporal correlated dataset into a hierarchical attention fusion network, and perform feature encoding on the data of each modality respectively; dynamically weight and fuse the features of each modality through an attention mechanism, and output a unified multi-modal feature representation combining the feature encoding; identify the anomaly points deviating from the normal pattern through an anomaly detection algorithm.

4. The humanoid robot joint module performance test method according to claim 3, wherein, The step of performing feature extraction and anomaly pattern detection on the multi-modal performance data through a cloud-pretrained multi-modal fusion model includes: compare the anomaly points with the ideal performance data simulated by the digital twin model, calculate a multi-dimensional deviation vector, and calibrate the parameters of the digital twin model based on the multi-dimensional deviation vector.

5. A humanoid robot joint module performance testing device, characterized by, The humanoid robot joint module performance testing device includes: a model loading module for obtaining the identity of the joint module under test, and loading the corresponding standard test procedure and digital twin model from the cloud based on the identity; a data acquisition module for controlling the test platform to perform an automated test sequence based on the standard test procedure, synchronously acquiring multi-modal performance data of the joint module under test through multiple sensors, and recording corresponding timestamps, specifically including: synchronously measuring the current harmonic data of the joint module under test through a high-precision electrical parameter analyzer; measuring the positioning error data of the joint module under test through a high-resolution optical encoder; collecting the vibration signal of the housing of the joint module under test through a high-frequency vibration accelerometer; collecting the audio signal of the joint module under test during operation through an acoustic microphone or ultrasonic sensor; and non-contact collecting the temperature field distribution data of the joint module under test through a thermal imaging camera; a data processing module for running a lightweight AI model through an edge computing unit to perform preliminary analysis on the multi-modal performance data, and adjusting the test strategy according to the preliminary analysis results, specifically including: time-aligning the synchronously acquired current harmonic data, positioning error data, vibration signal, audio signal, and temperature field distribution data; dynamically assigning weights to the multi-modal performance data using a lightweight AI model based on an attention mechanism, and fusing to generate a real-time health state vector; comparing and learning the real-time health state vector with the standard health state vector space of the digital twin model, calculating the relative distance or likelihood probability of the real-time health state vector in the standard health state vector space; if the relative distance exceeds a preset distance threshold or the likelihood probability is lower than a preset probability threshold, it is determined that there is an anomaly, and one or more preliminary hypothesis diagnoses are generated based on the deviation of each component in the real-time health state vector; inputting the preliminary hypothesis diagnosis and a current test procedure context into a test strategy generator pre-trained based on a meta-learning framework; dynamically selecting and combining a verification test sequence from a parameterized atomic test instruction library based on the test strategy generator; inserting the verification test sequence into the current test procedure and executing the verification test sequence to verify whether the preliminary hypothesis diagnosis is correct; if the preliminary hypothesis diagnosis is correct, generating a high-priority alarm, triggering a convergence test strategy, and constructing a triple based on the preliminary hypothesis diagnosis, the executed verification test sequence, and a verification result, and uploading the triple to a cloud to optimize and update the test strategy generator; if the preliminary hypothesis diagnosis is not correct, returning to the standard test procedure to continue execution; a cloud processing module configured to upload multi-modal performance data with an identity and a time stamp to a cloud data lake, perform deep analysis on the multi-modal performance data by an AI analysis engine of the cloud, and generate a test report.

6. A performance testing device for a humanoid robot joint module, characterized in that, The device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the anthropomorphic robot joint module performance test method according to any one of claims 1 to 4. The device comprises a memory, a processor, and a computer program stored on the memory and executable on the processor, and the computer program is configured to implement the steps of the anthropomorphic robot joint module performance test method according to any one of claims 1 to 4.

Images