Exoskeleton hierarchical safety detection method and device with abnormal backtracking

Abstract

The application relates to the technical field of exoskeleton control, and specifically discloses an exoskeleton hierarchical safety detection method and device with abnormal backtracking, which first classifies motion modes and extracts normalized features from a plurality of original data streams of an exoskeleton, then performs constraint consistency evaluation on the normalized feature vectors based on a working condition constraint parameter set to obtain an index set and an out-of-bound value set, further performs structural matching and backtracking on the index set by violating mode mapping rules to obtain an abnormal root cause category, simultaneously performs violation quantification and threshold matching on the out-of-bound value set to obtain a danger hierarchical grade, finally encapsulates the hierarchical strategy, violation degree weight and root cause warning information into a multi-modal intervention instruction package at the same frequency and distributes the multi-modal intervention instruction package for execution, drives a joint driver to implement slow-varying torque regulation and synchronously outputs root cause warning broadcasting. In this way, accurate determination and root cause backtracking of abnormalities are realized, and gradient slow-varying intervention is used to replace rough shutdown, so that the risk of imbalance and falling of a wearer is effectively reduced.

Description

Technical Field

[0001] This application relates to the field of exoskeleton control technology, and more specifically, to a graded safety detection method and device for exoskeletons with anomaly backtracking. Background Technology

[0002] Exoskeleton robots play a crucial role in assisting human movement in scenarios such as mobility rehabilitation and industrial load-bearing. The reliability of their safety detection systems directly impacts the wearer's safety. Current exoskeleton safety detection technologies generally employ a single-dimensional error threshold comparison method in anomaly detection. This means that the system's abnormality is determined solely by whether a certain type of sensor signal exceeds a fixed threshold. This logic cannot effectively distinguish between various easily confused scenarios such as normal gait fluctuations, loose clothing, sensor drift, and external disturbances, leading to frequent misjudgments and missed detections. More critically, when an anomaly is triggered, existing solutions can only identify its occurrence but cannot pinpoint and trace its specific source. The system lacks the ability to map constraint-violated structural features to specific root causes of failures, limiting subsequent handling to a uniform response strategy and hindering targeted, tiered intervention.

[0003] Patent application CN118744441A discloses an exoskeleton safety detection scheme. This scheme employs a single sensor data acquisition and fixed threshold judgment approach, directly executing a shutdown operation upon detecting an anomaly. However, this scheme lacks a multi-feature constraint consistency judgment mechanism, making it unable to accurately identify anomalies through the collaborative evaluation of multi-source feature vectors and multiple types of physical constraint functions. Furthermore, it lacks structured tracing capabilities based on constraint violation sets, failing to automatically distinguish between wear anomalies, control anomalies, external disturbance anomalies, and sensor anomalies. Moreover, its direct shutdown approach is highly susceptible to causing imbalance and falls when the wearer is walking or carrying weight, posing a significant risk of secondary injury. This makes it difficult to meet the comprehensive requirements of exoskeletons for safety detection accuracy, anomaly interpretability, and tiered protection under multiple motion conditions. Summary of the Invention

[0004] To address the aforementioned technical problems, this application is proposed. Embodiments of this application provide a method and apparatus for graded safety detection of exoskeletons with anomaly backtracking.

[0005] According to one aspect of this application, a graded safety detection method for exoskeletons with anomaly backtracking is provided, comprising: Step 1: Perform motion pattern classification and feature normalization extraction on the collected exoskeleton multi-source raw data stream to obtain the set of working condition constraint parameters and the exoskeleton normalized feature vector; Step 2: Based on the set of working condition constraint parameters, perform constraint consistency evaluation on the normalized feature vector of the exoskeleton to obtain the constraint violation index set and the constraint out-of-bounds value set. Step 3: Based on the preset violation pattern mapping rules, perform structural matching and source tracing on the constraint violation index set to obtain the anomaly root cause category; Step 4: Quantify the set of constraint violation values ​​and match them with threshold intervals to obtain the constraint violation degree value and the anomaly risk level. Step 5: Select the corresponding gradient intervention strategy based on the abnormal danger level, use the constraint violation value as the intensity adjustment weight, and retrieve the corresponding obstacle removal warning information based on the abnormal root cause category. Encapsulate and integrate the three in the same frequency to obtain a multimodal intervention execution instruction package. Step six: Parse and distribute the multimodal intervention execution command package to drive the exoskeleton joint actuator to implement gradual torque regulation and simultaneously drive the audio-visual interaction module to output root cause warning broadcast.

[0006] According to another aspect of this application, an exoskeleton graded safety detection device with anomaly backtracking is provided for the aforementioned exoskeleton graded safety detection method with anomaly backtracking, comprising: The feature normalization module is used to classify motion patterns and extract features from the acquired multi-source raw data stream of the exoskeleton, so as to obtain the set of working condition constraint parameters and the normalized feature vector of the exoskeleton. The constraint evaluation module is used to evaluate the constraint consistency of the exoskeleton normalized feature vector based on the set of working condition constraint parameters to obtain the constraint violation index set and the constraint out-of-bounds value set. The anomaly tracing module is used to perform structural matching and tracing of the constraint violation index set based on preset violation pattern mapping rules to obtain the anomaly root cause category. The risk classification module is used to quantify the set of constraint violation values ​​and match them with threshold intervals to obtain the constraint violation degree value and the abnormal risk level. The strategy encapsulation module is used to select the corresponding gradient intervention strategy based on the level of abnormal danger, use the constraint violation value as the intensity adjustment weight, and retrieve the corresponding troubleshooting warning information based on the abnormal root cause category. The three are encapsulated and integrated in the same frequency to obtain a multimodal intervention execution instruction package. The instruction execution module is used to parse and distribute multimodal intervention instruction packages for execution to drive the exoskeleton joint actuators to implement gradual torque control and simultaneously drive the audio-visual interaction module to output root cause warning broadcasts.

[0007] Compared with existing technologies, this application provides a hierarchical safety detection method and device for exoskeletons with anomaly backtracking. First, it employs multiple types of physical constraint functions to perform structured consistency evaluation on normalized feature vectors, effectively distinguishing various easily confused scenarios such as normal gait fluctuations, loose clothing, sensor drift, and external disturbances. This significantly reduces the false positive and false negative rates compared to single-dimensional threshold judgment. Second, through a structure matching and source tracing mechanism based on constraint violation index sets, the system can automatically locate the specific root cause category of the anomaly and output corresponding obstacle removal warning information, upgrading safety detection from traditional black-box alarms. For white-box traceability diagnosis, the targeted handling of anomalies and maintenance efficiency are improved. Then, the graded gradual intervention strategy uses the degree of violation value as the force adjustment weight to achieve fine torque control, avoiding the risk of secondary imbalance and fall caused by sudden shutdown. At the same time, the root cause broadcast output by the audio-visual module enables the wearer to know the cause of the anomaly in time and cooperate with the handling. Finally, the motion mode classification and constraint parameter dynamic matching mechanism enable the system to adaptively adjust the judgment sensitivity and tolerance range for different motion scenarios such as walking on flat ground, going up and down stairs, squatting and ramps, improving the detection stability and judgment accuracy in all scenarios. Attached Figure Description

[0008] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0009] Figure 1 This is a schematic diagram of the structure of an exoskeleton system according to an embodiment of this application; Figure 2 This is a flowchart of a graded safety detection method for exoskeletons with anomaly backtracking according to an embodiment of this application; Figure 3 This is a schematic diagram of the data flow of the exoskeleton graded safety detection method with anomaly backtracking according to an embodiment of this application; Figure 4 This is a flowchart of step five of the exoskeleton graded safety detection method with anomaly backtracking according to an embodiment of this application; Figure 5 This is a block diagram of an exoskeleton graded safety detection device with anomaly backtracking according to an embodiment of this application; Figure 6 This is a schematic diagram of the composition structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0010] To further illustrate the technical means and effects adopted by this application in order to achieve the intended purpose, the following detailed description of the specific implementation methods, structures, features and effects of this application is provided in conjunction with the accompanying drawings and preferred embodiments.

[0011] The technical solution of this application proposes a graded safety detection method for exoskeletons with anomaly backtracking. Figure 1 This is a schematic diagram of the structure of an exoskeleton system according to an embodiment of this application. Figure 1 As shown, the exoskeleton system achieves wear coupling with human limbs through flexible restraint components such as straps and buckles. The system includes an exoskeleton mechanical body, joint drive motors, joint encoders, inertial measurement units, strap tension sensors, buckle pressure sensors, and an audio-visual interaction module. Among them, multi-source sensors are used to synchronously collect human-machine coupling sensing data, joint drive motors receive control commands to implement torque regulation, and audio-visual interaction modules are used to output root cause warning broadcasts.

[0012] like Figure 2 and Figure 3 As shown, the exoskeleton graded safety detection method with anomaly backtracking according to an embodiment of this application includes the following steps: Step one involves classifying and normalizing the collected multi-source raw data streams of the exoskeleton to extract motion patterns, resulting in a set of constraint parameters for each working condition and a normalized feature vector for the exoskeleton. It's understandable that during actual wear and operation, the exoskeleton exhibits significantly different kinematic and dynamic characteristics in different motion conditions, such as walking on flat ground, climbing stairs, squatting, and navigating slopes. Using the same set of judgment parameters for all conditions would lead to inadequate detection sensitivity and tolerance ranges, resulting in misjudgments or missed judgments. Furthermore, the multi-source sensor data collected by the exoskeleton system is heterogeneous, with significant differences in the physical dimensions and numerical ranges of data from different channels. Directly incorporating this data into subsequent constraint evaluation without normalization would cause some large-range channels to exert unreasonable dominant biases on the judgment results. Therefore, this step classifies motion patterns to match specific constraint parameters for the current working condition and eliminates dimensional differences through feature normalization, providing fair and accurate data input for the constraint consistency evaluation in subsequent steps.

[0013] Specifically, in the embodiments of this application, step one includes: step 1.1, responding to the trigger signal of the exoskeleton closed-loop feedback control cycle, decoupling and distributing the multi-source raw data stream of the exoskeleton to obtain the raw motion posture data and the raw feature deviation and state data; step 1.2, determining the working condition threshold interval and mapping the parameter library based on the classification rules to obtain the working condition constraint parameter set; step 1.3, extracting the error scalar and normalizing the raw feature deviation and state data to obtain the normalized feature vector of the exoskeleton.

[0014] First, in step 1.1, when the system detects the start signal of a new control timing window, a high-frequency sampling interrupt is triggered, loading the multi-source raw data stream of the exoskeleton received within the current time window into a circular buffer. Then, channel decoupling is performed according to the preset data frame protocol header, extracting data frames used for attitude recognition, such as torso pitch angle signals, torso roll angle signals, and joint angle signals, and packaging them to generate raw motion attitude data. Simultaneously, data frames used for calculating abnormal deviations, such as joint angular velocity signals, joint angular acceleration signals, gait phase signals, wear coupling degree indicators, dynamic error judgment domains, and sensor communication quality signals, are extracted and packaged to generate raw feature deviation and state data. The purpose of this sub-step is to branch the mixed data stream according to its subsequent processing purpose, enabling motion pattern classification and feature extraction to process their respective required data subsets independently, either in parallel or serially.

[0015] Then, step 1.2 is executed. This sub-step uses a rule-based motion pattern classification method. By judging whether the joint angles and trunk pitch angles in the original motion posture data fall within the preset working condition threshold range, the discrete motion state label of the current exoskeleton is calculated. The specific recognition rules are as follows: when the trunk pitch angle is in the range of -5° to 5° and the joint angle is in the range of 30° to 60°, it is identified as a flat ground walking mode; when the trunk pitch angle is in the range of 7° to 13° and the joint angle is in the range of 40° to 90°, it is identified as a stair climbing mode; when the trunk pitch angle is in the range of -10° to -6° or 6° to 10°, it is identified as a ramp mode; when the joint angle is greater than 90° and the trunk forward tilt angle is greater than 10°, it is identified as a squatting mode; when the joint angular velocity is less than 0.1 rad / s, it is identified as a standing mode.

[0016] After calculating the motion state label, the label is used as an addressing index to perform a key-value pair lookup in a pre-set parameter mapping dictionary in the system's non-volatile memory. This extracts parameters such as sensitivity coefficients, tolerance threshold ranges, and allowable errors that strictly match the current working condition, generating a structured set of working condition constraint parameters. This parameter mapping dictionary pre-calibrates differentiated constraint parameters for each motion mode. For example, the speed error tolerance range for the flat ground walking mode can be set to 0 to 0.15, while the speed error tolerance range for the stair climbing mode can be set to 0 to 0.25, to adapt to the inherent differences in motion amplitude under different working conditions. The logical mapping process of this sub-step can be expressed as follows: first, the original motion posture data is mapped to motion category labels using a rule-based classification function; then, the motion category labels are mapped to the set of working condition constraint parameters using a parameter mapping dictionary function.

[0017] Finally, step 1.3 is executed. This sub-step first analyzes the feature deviation and original state data, extracts the absolute value of the difference between the actual operating value and the expected planned value of the exoskeleton, and obtains the original feature error scalars for each dimension. The extracted key error features include: velocity error, i.e., the deviation between the actual movement speed and the expected speed of the exoskeleton; acceleration error, i.e., the deviation between the actual movement acceleration and the expected acceleration of the exoskeleton; posture deviation, i.e., the deviation between the actual posture and the expected posture of the human body; phase difference, i.e., the deviation between the actual gait phase and the expected gait phase; and wear coupling index, a quantitative value characterizing the tightness of human-machine coupling. Among them, the wear coupling index is a positive representation quantity, and its larger value indicates a better coupling state. In the normalization step, the original values ​​are directly mapped dimensionlessly without taking the error difference value.

[0018] Subsequently, a minimum-maximum linear transformation algorithm is used to normalize each original feature error scalar, mapping them uniformly to a dimensionless interval of 0 to 1 to eliminate numerical abrupt interference caused by the physical dimensions of heterogeneous sensors. The normalization calculation formula is: the normalized eigenvalue of the k-th dimension is equal to the original feature error scalar of that dimension minus the theoretical minimum value of that dimension, divided by the difference between the theoretical maximum value and the theoretical minimum value of that dimension. The theoretical minimum value is usually taken as 0 or the value corresponding to the system's white noise floor, and the theoretical maximum value is the maximum error value achievable by the sensor's range or mechanical limits in that dimension. For example, for the speed error dimension, if the maximum allowable speed deviation of this model of exoskeleton is 2.0 m / s and the minimum deviation floor is 0.01 m / s, then when the actual speed error is 0.8 m / s, the normalized eigenvalue is (0.8-0.01) / (2.0-0.01) = 0.397.

[0019] After completing the normalization calculations for all dimensions, all m normalized feature values ​​are concatenated into column vectors according to a fixed dimensional order to generate a normalized feature vector for the exoskeleton. The dimension m of this vector is not less than 5, depending on the number of feature channels configured in the system. The concatenation order is as follows: normalized value of velocity error, normalized value of acceleration error, normalized value of attitude offset, normalized value of phase difference, normalized value of wear coupling degree, and normalized values ​​of other feature parameters supplemented according to actual needs.

[0020] In one specific embodiment, the exoskeleton wearer is currently walking on flat ground. After decoupling the multi-source raw data streams collected by the system, the raw motion posture data shows a torso pitch angle of 2° and a joint angle of 45°. Based on this, the current motion mode is determined to be walking on flat ground through rule classification. The system extracts the work condition constraint parameter set corresponding to walking on flat ground from the parameter mapping dictionary, where the speed error sensitivity coefficient is 1.2, the upper bound of the tolerance threshold is 0.15, and the allowable error is 0.05. Simultaneously, the feature deviation and state raw data show that the current speed error is 0.12 m / s and the acceleration error is... The orientation offset is 1.5°, the phase difference is 0.03 rad, and the wear coupling index is 0.85. After normalization, the feature values ​​of each dimension are mapped to 0.060, 0.040, 0.075, 0.030, and 0.850, respectively. These are then concatenated to generate a transposed column vector of the exoskeleton normalized feature vector [0.060, 0.040, 0.075, 0.030, 0.850].

[0021] Step two involves evaluating the constraint consistency of the exoskeleton's normalized feature vector based on the set of working condition constraint parameters to obtain a set of constraint violation indices and a set of constraint out-of-bounds values. It's understandable that traditional exoskeleton safety detection solutions often use single-dimensional threshold comparisons to determine anomalies, triggering alarms solely based on whether a single feature such as speed error exceeds limits. This fails to distinguish between fundamentally different anomaly scenarios such as normal gait fluctuations, loose clothing, sensor drift, and external disturbances, leading to high false positive and false negative rates. This step constructs multiple types of physical constraint functions and substitutes the normalized feature vectors for a structured consistency evaluation. This allows for simultaneous determination of whether the exoskeleton's operating state deviates from the safety envelope from multiple dimensions, including motion consistency, human-machine coupling, sensor consistency, and dynamic rationality, and accurately records which constraint dimensions have been violated and the specific magnitude of the violation.

[0022] Specifically, in this embodiment, step two includes: step 2.1, injecting the tolerance threshold and sensitivity coefficient of the working condition constraint parameter set into a preset basic human-machine physical constraint model to obtain a multidimensional instantiated constraint function cluster; step 2.2, inputting the exoskeleton normalized feature vector as an independent variable into the multidimensional instantiated constraint function cluster and performing parallel residual analysis on a class-by-class basis to obtain a constraint consistency evaluation quantization matrix; step 2.3, performing out-of-bounds logic screening and numerical index separation on the constraint consistency evaluation quantization matrix based on the zero-value threshold judgment criterion, wherein the out-of-bounds dimension numbers are aggregated to obtain a constraint violation index set, and the out-of-bounds magnitude values ​​are packaged to obtain a constraint out-of-bounds value set.

[0023] First, step 2.1 is executed. This sub-step first parses the set of working condition constraint parameters output from step one, extracting dynamic constraint scalars such as the tolerance threshold range, sensitivity coefficient, and allowable error corresponding to the current motion mode. Then, these dynamic constraint scalars are used as prior parameters and injected into the four types of basic human-machine physical constraint models preset by the system for model instantiation. The mathematical expression for instantiation is:

[0024] in, For the instantiated i-th type of security constraint function, it is the mathematical benchmark for evaluating whether the feature vector has exceeded the bounds; The equation skeleton of the i-th type of basic human-machine physical constraint model solidified in the system; Inputs are used for the multidimensional independent variables in the model, namely the normalized feature vector of the exoskeleton; The scalar of specific dynamic constraints for the i-th physical constraint model is extracted from the set of working condition constraint parameters. It includes parameters such as sensitivity coefficient, upper bound of tolerance threshold and allowable error. The value of i ranges from 1 to 4, where 1 represents motion consistency constraint, 2 represents human-machine coupling constraint, 3 represents sensor consistency constraint, and 4 represents dynamic rationality constraint.

[0025] Specifically, for the motion consistency constraint in the flat ground walking mode, the injected sensitivity coefficient can be 1.2, the upper limit of the tolerance threshold can be 0.15, and the allowable error can be 0.05; while for the same constraint in the stair-climbing mode, the injected sensitivity coefficient can be 0.8, the upper limit of the tolerance threshold can be 0.25, and the allowable error can be 0.08. For human-machine coupling constraints, the lower limit threshold of the wearable coupling degree in the flat ground walking mode can be set to 0.6, while it can be set to 0.5 in the squatting mode, to adapt to the inherent fluctuation range of human-machine coupling degree under different working conditions. Through parameter injection, the four types of basic constraint models are instantiated into a multi-dimensional instantiated constraint function cluster with a defined evaluation boundary, and the decision boundary of this function cluster is dynamically adjusted according to the working conditions.

[0026] Next, step 2.2 is executed. This sub-step treats the normalized exoskeleton feature vector output from step one as the independent variable and substitutes it into each type of constraint function in the multidimensional instantiation constraint function cluster generated in the previous sub-step. The system performs parallel matrix algebra operations to independently evaluate the deviation between the actual operating state of the feature vector under the four types of constraints and the theoretical safety boundary. The calculation result of each type of constraint function is a scalar. When the scalar is less than or equal to zero, it indicates that the feature vector is within the safety region in that constraint dimension; when the scalar is greater than zero, it indicates that the feature vector has exceeded the safety boundary in that constraint dimension, resulting in a constraint violation.

[0027] The calculation results of the four constraint functions are concatenated according to the constraint dimensions (1st to 4th dimensions) to generate a constraint consistency evaluation quantification matrix. The mathematical expression of this matrix is:

[0028] in, To obtain the constraint consistency evaluation quantization matrix, the real-time evaluation status of four types of constraints was recorded. The input exoskeleton normalized feature vector; to These represent the deviation scalars calculated after substituting the feature vector into the four instantiated constraint functions. Positive numbers indicate out-of-bounds errors, while negative numbers or zero indicate that the values ​​are within the safe region. It is a matrix transpose operator that combines multiple calculated scalars into a column vector.

[0029] Finally, step 2.3 is executed. This sub-step iterates through the values ​​of each dimension in the constraint consistency evaluation quantization matrix and performs a non-linear Boolean decision based on a threshold of zero. For the i-th component in the matrix, if its value is greater than 0, it is determined that the dimension has exceeded the constraint limit; if its value is less than or equal to 0, it is determined that the dimension is within the safe region.

[0030] After completing the judgment of all four dimensions, the system performs a two-way parallel data separation operation. The first path extracts the dimension numbers of all components with values ​​greater than zero and aggregates them to form a constraint violation index set. The second path simultaneously extracts the specific deviation values ​​with values ​​greater than zero in the matrix and packages them to form a constraint out-of-bounds value set. The mathematical expression for out-of-bounds logic screening and separation is:

[0031]

[0032] in, To prevent constraints from violating the index set, only the constraint number where the exception occurred is recorded; To constrain the set of out-of-bounds values, accurately record the magnitude of the violation of the constraint; The value of the i-th element in the constraint consistency evaluation quantization matrix. .

[0033] By separating the dimension number and the out-of-bounds magnitude, the constraint violation index set is used to support the qualitative tracing of the root causes of anomalies in step three, while the constraint out-of-bounds value set is used to support the quantitative calculation of the violation degree and risk classification in step four. If all constraint functions are satisfied, i.e., all components in the constraint consistency evaluation quantification matrix are less than or equal to zero, then the constraint violation index set is empty, and the constraint out-of-bounds value set is also empty. The exoskeleton is determined to be in normal operating condition, and the system directly outputs the normal state result without proceeding to the subsequent tracing and classification steps. If at least one constraint function is not satisfied, i.e., at least one component is greater than zero, then the exoskeleton is determined to be in an abnormal state, and the constraint violation index set and constraint out-of-bounds value set will continue to flow as input for subsequent steps.

[0034] In the first embodiment, it is assumed that the current exoskeleton wearer is walking on flat ground and the straps are gradually loosening. The set of work condition constraint parameters has been injected into four types of constraint models and instantiated. The normalized feature vector of the exoskeleton is... Substituting this feature vector into the four types of instantiation constraint functions respectively, the constraint consistency evaluation quantization matrix is ​​obtained as follows: The calculation result of the first dimension motion consistency constraint is -0.03, which is less than zero, indicating that the motion state is normal; the calculation result of the second dimension human-machine coupling constraint is 0.12, which is greater than zero, indicating that the wearer coupling degree has decreased, resulting in the violation of the human-machine coupling constraint; the calculation result of the third dimension sensor consistency constraint is -0.05, which is less than zero, indicating that the sensor state is normal; and the calculation result of the fourth dimension dynamic rationality constraint is -0.02, which is less than zero, indicating that the dynamic state is normal.

[0035] After boundary violation screening, only the second-dimensional component is greater than zero; therefore, the constraint violation index set is... This indicates that only the human-machine coupling constraint was violated. The set of constraint out-of-bounds values ​​is... The record shows that the out-of-bounds magnitude of the human-machine coupling constraint is 0.12. This constraint violates the index set. In step three, the set of constraint out-of-bounds values ​​will be matched to the wearable anomaly category by violating the pattern mapping rule. It will be used in the cumulative quantization calculation of the violation degree in step four.

[0036] In the second embodiment, it is assumed that the exoskeleton wearer experiences an external shove during walking, causing a sharp increase in the normalized values ​​of velocity and acceleration errors in the normalized feature vector of the exoskeleton output in step one. Substituting this feature vector into the four types of instantiation constraint functions yields the constraint consistency evaluation quantization matrix. The calculation result of the first dimension motion consistency constraint is 0.25, which is greater than zero, indicating that the motion state deviates from the normal gait pattern; the calculation result of the second dimension human-machine coupling constraint is -0.01, which is less than zero, indicating that the wearing state is normal; the calculation result of the third dimension sensor consistency constraint is -0.03, which is less than zero, indicating that the sensor state is normal; and the calculation result of the fourth dimension dynamic rationality constraint is 0.18, which is greater than zero, indicating that the mechanical equilibrium state has been broken.

[0037] After boundary violation logic screening, the first and fourth dimension components are greater than zero, therefore the constraint is violated and the index set is... This indicates that both the motion consistency constraint and the dynamic rationality constraint are violated simultaneously. The set of constraint out-of-bounds values ​​is as follows: This records the out-of-bounds magnitude of the two violation dimensions. This constraint violation index set... The set of constraint out-of-bounds values ​​that are matched to the external disturbance anomaly category by violating the pattern mapping rules in step three. In step four, the cumulative quantitative calculation of the degree of violation and the determination of the risk classification will be carried out.

[0038] Step three involves performing structural matching and source tracing on the constraint violation index set based on preset violation pattern mapping rules to obtain the root cause category of the anomaly. It should be understood that during the actual operation of the exoskeleton, different types of abnormal events can lead to violations of different combinations of constraint dimensions. For example, loosening of the garment only results in a violation of the human-machine coupling constraint, while external pushing simultaneously leads to a joint violation of motion consistency and dynamic rationality constraints. If the system can only detect the occurrence of an anomaly but cannot pinpoint its specific source, subsequent handling will lack targeted basis, and only a unified response strategy can be adopted, failing to provide effective troubleshooting guidance to the wearer. This step automatically translates the abstract number combinations into concrete physical fault categories by comparing the structural characteristics of the constraint violation index set with preset mapping rules.

[0039] Specifically, in this embodiment, step three includes: performing discrete subset structure comparison and anomaly label extraction on the constraint violation index set through the source hash mapping table to obtain the anomaly source root cause category.

[0040] Specifically, the dimension-numbered elements contained in the set are first read and constructed into a discrete mathematical subset structure. For example, if the constraint violation index set output in step two is {2}, the constructed subset structure is a set containing only the single element 2; if the constraint violation index set is {1, 4}, the constructed subset structure is a combined set containing elements 1 and 4.

[0041] Subsequently, the system uses the constructed mathematical subset structure as the retrieval key and performs addressing and comparison in the system's built-in source hash mapping table. The source hash mapping table pre-stores the mapping relationships between all violation patterns and anomaly sources, identified based on the physical laws of exoskeleton operation and engineering experience. The specific mapping rules are as follows.

[0042] When the constraint violation index set is {2}, meaning only the human-machine coupling constraint is violated, the anomaly category output by the mapping table is "wearing anomaly." The corresponding physical fault scenarios include loose straps and wearable displacement. The backtracking basis is the changing trend of the wearable coupling degree index and the out-of-bounds situation of the dynamic error judgment domain. The physical logic of this mapping relationship is that: loose wear directly leads to a decrease in human-machine coupling degree, causing the calculation result of the human-machine coupling constraint function to break through the zero boundary. At this time, the exoskeleton's own motion control, sensor status, and dynamic characteristics are not affected, so the other three types of constraint functions remain satisfied.

[0043] When the constraint violation index set is {1}, meaning only the motion consistency constraint is violated, the anomaly category output by the mapping table is control anomaly. The corresponding physical fault scenarios include excessive or insufficient assist output, excessive motion tracking error, etc. The backtracking basis is the temporal consistency of velocity error, acceleration error, and phase difference. The physical logic of this mapping relationship is that the deviation of the control algorithm output causes the actual motion trajectory of the exoskeleton to deviate from the expected trajectory, resulting in the violation of the motion consistency constraint, but the human-machine coupling state, sensor hardware, and mechanical balance are not affected.

[0044] When the constraint violation index set is {1, 4}, meaning both motion consistency constraints and dynamic rationality constraints are violated simultaneously, the anomaly category output by the mapping table is external disturbance anomaly. The corresponding physical fault scenarios include external force impacts such as tripping, collisions, and pushing. The backtracking basis is the abrupt change characteristics of velocity and attitude, as well as phase mismatch. The physical logic of this mapping relationship is that when a sudden external force acts on the wearer, it will inevitably cause a drastic change in motion velocity and acceleration (violating motion consistency constraints) and break the mechanical equilibrium state (violating dynamic rationality constraints). However, the wearer's state and the sensor hardware itself do not change, so the human-machine coupling constraint and the sensor consistency constraint are still satisfied.

[0045] When the constraint violation index set is {3}, meaning only the sensor consistency constraint is violated, the anomaly category output by the mapping table is sensor anomaly. The corresponding physical fault scenarios include sensor drift, data packet loss, signal jumps, etc., and the backtracking basis is the cross-validation results of multi-sensor data and the sensor communication status. The physical logic of this mapping relationship is that a hardware failure of a single sensor causes its output data to be inconsistent with other sensor data, resulting in a violation of the sensor consistency constraint, but the actual motion state, wearing state, and dynamic characteristics of the exoskeleton do not change.

[0046] After completing the addressing and comparison of the hash mapping table, if the subset structure of the current constraint violation index set successfully matches a rule in the mapping table, the system extracts the anomaly category label bound to that rule, encodes and formats it, and generates the final output anomaly root cause category. The mathematical expression of this tracing process is: the anomaly root cause category equals the mapping output value of the source hash mapping function on the constraint violation index set. If the structure of the constraint violation index set fails to find an exact match in the mapping table, the system may use a nearest neighbor matching strategy or output an unknown anomaly category label, and record this unmatched event in the system log for subsequent manual analysis.

[0047] In the first embodiment, assuming the exoskeleton wearer's straps gradually loosen while walking on flat ground, the constraint violation index set output in step two is {2}. The system uses the subset structure {2} as a key to search the source hash mapping table, matching the mapping rule "key {2} corresponds to wearing anomaly", extracting the anomaly label as wearing anomaly and encapsulating it for output. The root cause category of the anomaly is wearing anomaly. This result indicates that the root cause of the current anomaly lies in the deterioration of the human-machine coupling state, specifically manifested as a decrease in the wearing coupling index due to loose straps. Based on this, the system can retrieve troubleshooting warning information for wearing looseness in subsequent steps.

[0048] In the second embodiment, assuming the exoskeleton wearer experiences a lateral shove during walking, the constraint violation index set output in step two is {1,4}. The system uses the subset structure {1,4} as a key to search the source hash mapping table, matching the mapping rule "key {1,4} corresponds to external disturbance anomaly". The anomaly label is extracted as external disturbance anomaly and encapsulated for output. The root cause category of the anomaly is external disturbance anomaly. This result indicates that the current anomaly is not due to a fault in the device itself, but rather caused by a sudden external impact. Based on this, the system can trigger a higher level of protection strategy and output corresponding safety warnings in subsequent steps.

[0049] Step four involves quantifying the set of constraint violation values ​​and matching them with threshold intervals to obtain constraint violation values ​​and anomaly risk levels. It's understandable that during the actual operation of the exoskeleton, subsequent graded intervention strategies require a clear level criterion to determine the appropriate level of protection. Without quantitative violation values ​​and standardized risk levels, the system cannot achieve refined gradient intervention and can only apply the same level of treatment to all anomalies. This step quantifies and accumulates the violation values ​​to obtain continuous violation values, and then matches these values ​​with preset stepped thresholds to obtain discrete risk levels.

[0050] Specifically, in this embodiment of the application, step four includes: step 4.1, performing cumulative calculation of all positive out-of-bounds values ​​in the constraint out-of-bounds value set and quantitative evaluation of the degree of violation to obtain the constraint violation value; step 4.2, performing interval definition and level matching of the constraint violation value based on a preset segmented step threshold to obtain the abnormal danger level.

[0051] First, execute step 4.1, which receives the set of constraint out-of-bounds values ​​output from step 2. The elements in this set are already pre-selected positive out-of-bounds residual values ​​that are strictly greater than zero.

[0052] In the first embodiment, for all positive out-of-bounds values ​​within the constraint out-of-bounds value set, a linear summation function is invoked to perform an accumulation operation, calculating a scalar value characterizing the total deviation of the exoskeleton from the safety envelope. The mathematical expression for this linear accumulation calculation is:

[0053] in, To constrain the degree of violation, it is a non-negative continuous real scalar; For the i-th type of constraint function, normalize the feature vector of the exoskeleton. The calculation results; This indicates that the larger value between the constraint function calculation result and zero is taken, ensuring that only the positive out-of-bounds portion is accumulated; the value of i ranges from 1 to 4, corresponding to motion consistency constraints, human-machine coupling constraints, sensor consistency constraints, and dynamic rationality constraints, respectively. This formula is equivalent to performing an arithmetic summation on all elements in the constraint out-of-bounds value set, that is:

[0054] in, To constrain the set of out-of-bounds values, This refers to the iterative elements in the set, i.e., the positive out-of-bounds residual scalar of a single dimension. This implementation has simple computational logic and is suitable for application scenarios where the physical coupling between constraint dimensions is weak or negligible.

[0055] Research shows that when an exoskeleton wearer experiences a sudden external impact (such as tripping while walking or being pushed laterally), both motion consistency constraints and dynamic rationality constraints inevitably exceed their limits simultaneously. This results in drastic changes in velocity and acceleration, and also disrupts the mechanical equilibrium. In this multi-dimensional joint limit-exceeding scenario, the actual risk of fall faced by the wearer far exceeds the arithmetic sum of the limit-exceeding magnitudes of each dimension, indicating a nonlinear, synergistic danger amplification effect between the two constraint dimensions. In contrast, when only the sensor consistency constraint exceeds its limit (such as signal drift in an encoder), it is an isolated hardware failure, not physically coupled with other constraint dimensions. Its threat to the wearer's safety is limited to the extent reflected by its own limit-exceeding magnitude.

[0056] The linear summation operator in the first embodiment produces the same violation value for both the coupled multidimensional joint boundary violation and the isolated single-dimensional boundary violation, making it impossible to distinguish the essential differences in their physical hazard levels. This leads to a systematic underestimation of the true hazard level in multidimensional coupled boundary violation scenarios, causing a lag bias in the subsequent determination of abnormal hazard levels, ultimately delaying the triggering of tiered intervention and increasing the probability of the wearer falling under sudden impact.

[0057] To address the aforementioned shortcomings, a second embodiment is proposed. This embodiment upgrades the original one-dimensional linear summation to a quadratic matrix operation by introducing a constraint coupling amplification weight matrix, enabling the constraint violation value to simultaneously encode the self-boundary magnitude of each dimension and the cross-dimensional coupling amplification contribution.

[0058] Specifically, in the second embodiment, step 4.1 includes: based on the constraint dimension attribution identifier carried by each element in the constraint out-of-bounds value set, reconstructing the constraint out-of-bounds value set in ascending order of dimension numbers and concatenating it into column vectors to obtain an ordered vector of out-of-bounds values; using the combination of dimension numbers corresponding to each component in the ordered vector of out-of-bounds values ​​as a two-dimensional index pair, extracting submatrix slices in a preset global coupling amplification prior matrix to obtain an active-state coupling amplification weight submatrix; and performing a quadratic coupling operation on the ordered vector of out-of-bounds values ​​based on the active-state coupling amplification weight submatrix to obtain the constraint violation degree value.

[0059] First, based on the constraint dimension attribution identifiers carried by each element in the constraint out-of-bounds value set, the constraint out-of-bounds value set is reconstructed in ascending order of dimension numbers and concatenated into column vectors to obtain an ordered vector of out-of-bounds values. Since the elements in the constraint out-of-bounds value set exist in a discrete and unordered form, they cannot directly participate in matrix algebra operations. Therefore, it is necessary to parse the constraint dimension attribution identifiers carried by each element in the set and reconstruct the discrete set into a column vector with a definite dimension order according to the ascending order of dimension numbers. The length of this vector is equal to the number of constraint dimensions that actually went out of bounds in the current detection period, and the position of each component strictly corresponds to its constraint dimension number, thus establishing a clear dimension index basis for subsequent matrix operations. For example, if the constraint out-of-bounds value set is {0.25, 0.18} and the corresponding dimension numbers are 1 and 4 respectively, then the reconstructed ordered vector of out-of-bounds values ​​is:

[0060] Its first component corresponds to dimension number 1, and its second component corresponds to dimension number 4.

[0061] Subsequently, using the dimension number combinations corresponding to each component in the ordered vector of out-of-bounds values ​​as two-dimensional index pairs, submatrix slicing is performed in the pre-stored global coupling amplification prior matrix in the system's non-volatile memory to obtain the activation-state coupling amplification weight submatrix. The global coupling amplification prior matrix is ​​a symmetric positive definite matrix with a dimension of 4x4, corresponding to all pairwise combinations of the four types of constraint dimensions, and its general form is:

[0062] Among them, diagonal elements Encode the self-weight coefficients of the i-th constraint dimension, off-diagonal elements Encode the physical coupling amplification factor between the i-th and j-th constraint dimensions, and satisfy symmetry. For example, the off-diagonal elements corresponding to motion consistency constraints (dimension 1) and dynamic rationality constraints (dimension 4). The value can be set to 1.8 to reflect the synergistic danger amplification effect when the two elements jointly cross the boundary; the off-diagonal elements corresponding to the sensor consistency constraint (dimension 3) and other dimensions can be set to 0.05 to reflect their physical independence. The diagonal elements can be set according to the weight of each dimension's impact on safety, for example... , , , Slicing extraction retains only the row and column intersection regions corresponding to the current activation dimension, generating an activation-state coupling amplification weight submatrix with an order equal to the number of current activation dimensions. For example, when the activation dimensions are {1, 4}, the extracted submatrix is:

[0063] Finally, a quadratic coupling operation is performed on the ordered vector of out-of-bounds values ​​based on the activated-state coupling amplification weight submatrix to obtain the constraint violation degree. The mathematical expression for this operation is:

[0064] in, To constrain the violation value, it is a non-negative continuous real scalar, representing the severity of the exoskeleton's overall deviation from the safe operating envelope at the current moment; A bounded ordered vector of values, with dimension . ( (This represents the number of currently active out-of-bounds constraint dimensions), and each component represents the positive out-of-bounds residual magnitude value of the corresponding constraint dimension; The transpose of the ordered vector of out-of-bounds values, with dimension . ; The activation-state coupling amplification weight submatrix has a dimension of . Its diagonal elements Represents the independent hazard weight coefficient of the i-th activation constraint dimension, and the off-diagonal elements. This represents the physical coupling amplification factor between the i-th and j-th activation constraint dimensions.

[0065] When only a single dimension exceeds the bounds, the ordered vector of the out-of-bounds value degenerates into a scalar. The activated-state coupling amplification weight submatrix degenerates into a single element. At this point, the constraint violation value degenerates to:

[0066] When the cross term is zero, the constraint violation value is contributed only by the weighted square of that dimension itself, equivalent to a weighted independent evaluation. When multiple dimensions jointly exceed the limit, expanding the quadratic form yields:

[0067] The first term on the right is the sum of the weighted squared contributions of each dimension, and the second term is the sum of the cross-coupling terms across dimensions. The off-diagonal coupling factor activates the cross-product term, causing the constraint violation degree to grow superlinearly, which can reflect the physical fact of the collaborative danger amplification under multi-dimensional joint boundary violation scenarios.

[0068] Finally, step 4.2 is executed. This sub-step extracts the constraint violation value calculated in the previous sub-step and injects it into a preset segmentation condition judgment function. It then compares this value with the stage safety threshold set in the system's non-volatile memory, matching the corresponding standardized numerical level based on the specific numerical range the value falls into. The mathematical expression for segmentation judgment is:

[0069] in, The output is the level of abnormal danger, and the value is the set of discrete natural numbers {0,1,2,3}. This is the constraint violation value calculated in the previous sub-step; To determine the upper bound threshold for minor anomalies that have been pre-calibrated experimentally; To determine the upper bound threshold for moderate anomalies, which is pre-calibrated experimentally, and which necessarily satisfies... .when When all constraint functions are satisfied, the exoskeleton operates normally; when When the time indicates a low level of abnormality, it does not affect basic operation; when When the time indicates a moderate level of abnormality, it is necessary to reduce the operating power to ensure safety; when This indicates a severe abnormality, requiring immediate and mandatory protective measures.

[0070] in, and The specific value can be determined based on different motion patterns, exoskeleton models, and experimental data. For example, in a scenario where the violation degree is calculated using the linear summation method of the first embodiment, , In scenarios where the violation degree is calculated using the quadratic form operation method of the second embodiment, the numerical range of the quadratic form operation differs from that of linear summation. , .

[0071] In the example of the first embodiment, assuming the exoskeleton wearer experiences a loosening of the straps while walking on flat ground, the constraint out-of-bounds value set output in step two is {0.12}, containing only the out-of-bounds magnitude of the human-machine coupling constraint. It is calculated using a linear summation method:

[0072] The value is compared with a preset threshold, because ,therefore The abnormality risk level is classified as Level 1, minor anomaly. As the wearable coupling continues to decrease, if the set of constraint out-of-bounds values ​​becomes {0.28} in subsequent detection cycles, then:

[0073] because The abnormal danger level has been upgraded to At that time, it was classified as a level 2 moderate abnormality.

[0074] In the example of the second embodiment, assuming the exoskeleton wearer experiences an external shove, the set of constraint out-of-bounds values ​​output in step two is {0.25, 0.18}, corresponding to dimension numbers 1 (motion consistency constraint) and 4 (dynamic rationality constraint), respectively. First, this set is reconstructed into an ordered vector of out-of-bounds values ​​in ascending order of dimension numbers:

[0075] Using the dimension number combination {1, 4} as a two-dimensional index pair, extract the activation-state coupling amplification weight submatrix from the global coupling amplification prior matrix:

[0076] Perform quadratic operations:

[0077] Expanding the calculation as follows:

[0078] The value is compared with the preset threshold corresponding to the second embodiment. The abnormal risk level is determined as follows: This indicates a level 2 moderate abnormality.

[0079] In contrast, if the linear summation method of the first embodiment is used for the same scenario, the constraint violation value is... This corresponds to a level 2 moderate anomaly under the threshold system of the first embodiment. The second embodiment, however, uses a coupling factor... The activation of the interaction term resulted in the interaction term contributing additional... Compared to the independent weighted sum of squares for each dimension, the superlinear growth resulting from the activation of coupled cross terms more accurately reflects the amplified collaborative danger faced by the wearer when the joint boundaries of motion consistency constraints and dynamic rationality constraints are exceeded. In practical applications, the calculation method of the first or second embodiment can be selected depending on the specific model of the exoskeleton and the application scenario.

[0080] Step 5 involves selecting a corresponding gradient intervention strategy based on the anomaly risk level, using the constraint violation value as the force adjustment weight, and retrieving corresponding troubleshooting warning information based on the anomaly root cause category. These three elements are then encapsulated and integrated in sync to obtain a multimodal intervention execution instruction package. It should be understood that during the actual operation of the exoskeleton, safety intervention needs to simultaneously meet two requirements: firstly, it requires applying torque control to the joint actuators that precisely matches the severity of the anomaly, avoiding insufficient protection due to too little intervention or imbalance due to too much; secondly, it requires synchronously outputting troubleshooting guidance corresponding to the root cause of the anomaly to the wearer, enabling them to understand the current situation and cooperate in handling it. If mechanical intervention and interactive warnings are not synchronized on the timeline, the wearer may experience changes in exoskeleton movement before knowing the cause of the anomaly, leading to panic, or may not feel protective action after receiving warnings, leading to distrust. This step ensures that mechanical intervention and human-machine interaction are triggered collaboratively at the same time by encapsulating the grading strategy, violation weight, and root cause warning information in sync.

[0081] Specifically, in the embodiments of this application, such as Figure 4 As shown, step five includes: Step 5.1, performing macro-intervention strategy matching based on the baseline strategy lookup table to obtain the baseline strategy, and then dynamically modulating the constraint violation degree value with the baseline strategy after converting it into control strength adjustment weight to obtain the hardware physical intervention control logic; Step 5.2, performing retrieval mapping based on the expert troubleshooting database with the anomaly root cause category as the primary key index to obtain specific object interaction warning data; Step 5.3, performing synchronous stamping and protocol encapsulation on the hardware physical intervention control logic and the specific object interaction warning data to obtain a multimodal intervention execution instruction package.

[0082] First, step 5.1 is executed. This sub-step first extracts the abnormal hazard level output from step four and retrieves the matching macro-intervention strategy from the preset benchmark strategy lookup table. The preset intervention schemes for each level in the benchmark strategy lookup table are as follows: When the abnormal hazard level is level 1, the corresponding benchmark strategy is torque limiting, with a benchmark limiting ratio range of 0.20 to 0.40; when the abnormal hazard level is level 2, the corresponding benchmark strategy is linear power reduction, with a benchmark power reduction target of reducing the operating power to 0.50 of the rated power and a power reduction transition time of 3 to 5 seconds; when the abnormal hazard level is level 3, the corresponding benchmark strategy is to stop the power assist output and perform joint easing and mechanical locking.

[0083] After completing the baseline strategy matching, the system extracts the constraint violation values ​​output from step four and inputs them as independent variables into a nonlinear scaling function to calculate the control intensity adjustment weight. The scaling function maps continuous violation values ​​to a dimensionless coefficient used to fine-tune the implementation intensity of the baseline strategy. The mathematical expression of the scaling function is:

[0084] in, Adjust the weight to control the intensity; To constrain the degree of violation; This is the lower bound threshold of the current level range; This is the upper limit threshold of the current level range; This represents the maximum value of the intensity adjustment weight within this level. This is the minimum value of the intensity adjustment weight within this level.

[0085] Subsequently, the baseline strategy and control strength adjustment weights are dynamically modulated to generate the final hardware physical intervention control logic. The mathematical expression for dynamic modulation is:

[0086] in, The generated hardware physical intervention control logic is represented as control parameters that include torque limiting ratio or power attenuation coefficient. The baseline strategy lookup function is used to classify discrete anomaly risk levels. Mapped to a standard intervention intensity benchmark value; The constraint violation value The control force adjustment weight is calculated using a scaling function. For example, when the abnormal hazard level is 1 and the baseline limit ratio is 0.20, if the control force adjustment weight is 0.9, the final torque limit ratio will be... If the baseline limiting ratio is 0.40 and the weight is 0.9, then the final limiting ratio will be... Through this dynamic modulation mechanism, different violation values ​​within the same level will generate different intervention forces, achieving a refined transition from discrete levels to continuous forces.

[0087] Next, step 5.2 is executed. This sub-step receives the anomaly root cause category output from step three, uses this category label as a unique primary key index, and performs a retrieval operation in the expert troubleshooting database built into the system's non-volatile memory. The expert troubleshooting database stores the mapping relationship between all labeled anomaly root cause categories and corresponding troubleshooting information in the form of key-value pairs. Each record contains system diagnostic scripts, recommended operation guidelines, and audio-visual control codes for that root cause.

[0088] The specific retrieval mapping relationships are as follows: When the root cause category of the anomaly is "wearing abnormality," the retrieved diagnostic message is "Loose strap, please tighten," and the recommended operation is to tighten the strap on the corresponding limb segment. The audible and visual control code is a low-frequency yellow flashing followed by a voice announcement. When the root cause category of the anomaly is "control abnormality," the retrieved diagnostic message is "Assist deviation, automatically adjusting," and the recommended operation is to maintain the current posture and wait for the system to self-correct. The audible and visual control code is a mid-frequency blue flashing followed by a voice announcement. When the root cause category of the anomaly is "external disturbance abnormality," the retrieved diagnostic message is "External impact detected, system protection activated," and the recommended operation is to hold the object firmly and wait for the system to recover. The audible and visual control code is a high-frequency red flashing followed by a buzzer alarm and a voice announcement. When the root cause category of the anomaly is "sensor abnormality," the retrieved diagnostic message is "Sensor signal abnormality, please check the equipment," and the recommended operation is to stop using the device and contact maintenance personnel. The audible and visual control code is a mid-frequency orange flashing followed by a voice announcement.

[0089] The system encodes and formats the retrieved diagnostic scripts, recommended operation guidelines, and audio-visual control codes to generate interactive alert data for specific users. The mathematical expression for this retrieval and mapping process is:

[0090] in, The interactive warning data for specific objects includes data structures such as speech-synthesized text address codes, LED flashing pattern words, and buzzer frequency parameters. A retrieval mapping operator for the expert troubleshooting database; This is a category label for tracing the root cause of anomalies.

[0091] Finally, step 5.3 is executed. This sub-step first extracts the high-precision real-time clock tick of the current system and generates a globally unified timestamp. Then, the hardware physical intervention control logic generated in the first sub-step and the specific object interaction warning data generated in the second sub-step are pulled into the same cache bus. The two heterogeneous data streams are then synchronously stamped and encapsulated using the timestamp, with checksums and frame header / tail instruction set protocols added, and merged to generate a multimodal intervention execution instruction package.

[0092] In the first embodiment, assuming the exoskeleton wearer experiences a loosening of the straps while walking on flat ground, the anomaly root cause category output in step three is "wearing abnormality," and the constraint violation value output in step four is 0.12, with an anomaly risk level of 1. The system first retrieves the baseline strategy from the baseline strategy lookup table using level 1 as the index, obtaining a baseline strategy of torque limiting with a baseline limiting ratio of 0.30. Then, the constraint violation value of 0.12 is input into the scaling function; for example, for a level 1 minor anomaly, , , , When the constraint violation value is 0.12, the calculated control strength adjustment weight is: The baseline limiting ratio and weights are modulated to obtain a final torque limiting ratio of 0.30 × 0.9 = 0.27, meaning the hardware physical intervention control logic limits the joint output torque to within 0.27 times the rated torque. Simultaneously, using the wearability abnormality as the primary key, a search is performed in the expert troubleshooting database to obtain specific object-specific interactive warning data: a voice broadcast saying "The strap is loose, please tighten it," accompanied by a low-frequency flashing yellow LED. Finally, the torque limiting control parameters and the voice broadcast data are encapsulated with the current timestamp to generate a multimodal intervention execution instruction package.

[0093] In the second example, assuming the exoskeleton wearer experiences an external shove, the anomaly root cause category output in step three is external disturbance anomaly, and the constraint violation value output in step four is 0.2731 (using the quadratic form calculation in the second embodiment), with an anomaly risk level of 2. The system retrieves the baseline strategy from the baseline strategy lookup table using level 2 as the index, finding it to be linear power reduction. The baseline power reduction target is to reduce the operating power to 0.50 of the rated power, with a transition time of 3 seconds. The constraint violation value of 0.2731 is input into the scaling function, using level 2 interval parameters. , , , The calculated control force adjustment weight is: The benchmark power reduction target is modulated with weights to obtain the final power attenuation coefficient. Specifically, the hardware physical intervention control logic linearly reduces the joint output power to 0.443 times the rated power within 3 seconds. Simultaneously, using external disturbance anomalies as the primary key, it searches the expert troubleshooting database to obtain specific object-specific interactive warning data: a voice broadcast stating "External impact detected, system protection activated," accompanied by high-frequency red LED flashing and a buzzer alarm. Finally, the power attenuation control parameters and the audible and visual alarm data are encapsulated with the current timestamp to generate a multimodal intervention execution command package.

[0094] Step six involves parsing and splitting the multimodal intervention execution command packet for execution, driving the exoskeleton joint actuators to implement gradual torque regulation and simultaneously driving the audio-visual interaction module to output root cause warning broadcasts. It should be understood that during actual exoskeleton operation, the joint servo actuators and the audio-visual interaction module belong to different communication buses and control domains, with differences in data frame formats, communication protocols, and response timings. If unparsed mixed command packets are directly sent, each terminal will be unable to correctly identify its own payload. Furthermore, if the joint actuators directly execute a step-like torque change after receiving the torque regulation command, it will cause a sudden change in the exoskeleton's output torque, potentially causing the wearer to lose balance and fall during movement due to sudden loss or gain of assistance. Therefore, the torque regulation command must be processed gradually to ensure a smooth transition in torque changes. This step achieves synchronous and coordinated execution of mechanical protection and human-machine interaction by parsing the command packet's protocol and splitting the channel, then performing signal conditioning and sending the two commands separately according to the characteristics of their respective terminals.

[0095] Specifically, in this embodiment, step six includes: step 6.1, performing verification, demultiplexing, and channel stripping on the multimodal intervention execution instruction packet based on a preset frame protocol, wherein the torque control frame is reconstructed into a downlink hardware drive instruction stream, and the interactive warning frame is reconstructed into a downlink interactive warning data stream; step 6.2, injecting the downlink hardware drive instruction stream into the joint servo driver for execution after performing a gradual smoothing filter, and simultaneously injecting the downlink interactive warning data stream into the audio-visual module to execute root cause broadcasting.

[0096] First, step 6.1 is executed. The execution process of this sub-step is as follows: The exoskeleton main control MCU receives the multimodal intervention execution instruction packet transmitted on the communication bus. First, it performs an integrity check on the instruction packet based on the preset CRC check polynomial. It compares the checksum field in the received data frame with the locally recalculated checksum value. If the two are consistent, it is determined that there are no transmission errors and the subsequent parsing process is allowed. If they are inconsistent, the frame is discarded and a retransmission is requested to prevent communication errors from causing incorrect intervention actions.

[0097] After successful verification, the system invokes the protocol demultiplexing algorithm to perform channel stripping on the payload of the command packet according to the preset frame header identifier field. Specifically, the system scans the payload area of ​​the command packet byte by byte. When the frame header identifier of a torque control frame is identified (e.g., the frame header identifier byte is 0xA1), the fixed-length data segment following the frame header is extracted as the torque control payload. When the frame header identifier of an interactive warning frame is identified (e.g., the frame header identifier byte is 0xB2), the fixed-length data segment following the frame header is extracted as the interactive warning payload. The extracted torque control payload is reconstructed according to the PDO message format of the CAN bus or EtherCAT bus to generate a downlink hardware driver command stream. The extracted interactive warning payload is reconstructed according to the data frame format of the serial port or I2C bus to generate a downlink interactive warning data stream.

[0098] Next, execute step 6.2, which includes two parallel operations: mechanical intervention and human-computer interaction.

[0099] In the mechanical intervention execution path, the system injects the downlink hardware drive command stream into the underlying joint servo driver. Upon receiving the torque control target value, the servo driver does not directly jump the output torque to the target value. Instead, it uses an integral ramp generator to perform gradual smoothing filtering on the torque change. The core of this gradual processing is to constrain the torque change rate with a preset slope limit value, so that the output torque transitions from the current actual value to the target value in a linear and gradual manner. The mathematical expression for gradual torque control is:

[0100] in, This represents the actual output torque value at time t. The current output torque value at the moment the intervention command is received; The target torque limit value specified in the downlink hardware driver instruction stream; This is a sign function used to determine the direction of torque change; The preset rate of change of torque ramp, in N·m / s; This is the moment when the intervention command begins to be executed. achieve During this process, the output torque remains at the target value and does not change. This gradual change mechanism ensures that when the exoskeleton performs torque limiting, power reduction, or slow-stop operations, the wearer can feel the gradual change in assistance and has time to adjust their posture to maintain balance, avoiding falls due to sudden torque changes.

[0101] The specific implementation of the gradual change process varies depending on the level of abnormality risk. When the abnormality risk level is Level 1, the servo drive performs a torque limiting operation, reducing the output torque in a ramp manner to the target value of the rated torque multiplied by the limiting ratio. The limiting ratio ranges from 0.20 to 0.40, and the transition time is determined by the ramp rate. When the abnormality risk level is Level 2, the servo drive performs a linear power reduction operation, linearly reducing the output power to 0.50 times the rated power within a 3-5 second transition time, while continuously monitoring the trend of constraint violation. When the abnormality risk level is Level 3, the servo drive first stops the assist output, then performs a joint slow-stop operation, smoothly reducing the joint angular velocity to zero at a preset deceleration slope, and finally triggers the mechanical locking mechanism to fix the joint position. The entire process avoids sudden shutdown that could cause the wearer to lose balance.

[0102] In the human-computer interaction execution path, the system synchronously injects the downlink interactive warning data stream into the audio-visual interaction module. The audio-visual interaction module includes a speech synthesis chip, an LED indicator controller, and a buzzer driver circuit. After receiving the text address code, the speech synthesis chip retrieves the corresponding speech segment from its built-in speech library and broadcasts the root cause of the abnormality and handling suggestions to the wearer through an external speaker. After receiving the flashing mode word, the LED indicator controller drives the indicator light to flash according to the specified color and frequency. After receiving the frequency parameter, the buzzer driver circuit drives the buzzer to sound at the specified frequency. Since both the downlink hardware drive command stream and the downlink interactive warning data stream carry the unified timestamp entered in step five, the two execution terminals synchronously trigger based on the timestamp after parsing their respective commands, ensuring that the wearer hears the voice broadcast and sees the indicator light change at the same moment they feel the torque change, achieving timing alignment between mechanical protection and information notification.

[0103] In the first embodiment, assuming the exoskeleton wearer experiences a loosening of the straps while walking on flat ground, the multimodal intervention execution command packet generated in step five includes a level 1 torque limiting control parameter (limiting ratio of 0.27) and voice broadcast data indicating a wearing abnormality. The system first performs a CRC check on the command packet to confirm it is error-free. Then, it extracts the torque control frame using frame header identifier 0xA1 and reconstructs it into a downlink hardware driver command stream in CAN bus format. It also extracts the interactive warning frame using frame header identifier 0xB2 and reconstructs it into a downlink interactive warning data stream in serial port format. After receiving the downlink hardware driver command stream, the joint servo driver executes it at a ramp rate. The output torque will be smoothly reduced from the current 10.0 N·m to the target value. The transition time is approximately Seconds. At the same moment the torque begins to decrease, after receiving the downlink interactive warning data stream, the voice synthesis chip announces "The strap is loose, please tighten it," and the LED indicator operates in a low-frequency yellow flashing mode, indicating to the wearer that the current abnormality is that the clothing is loose and the severity is low.

[0104] In the second embodiment, assuming the exoskeleton wearer experiences an external shove, the multimodal intervention execution command packet generated in step five includes two levels of linear power reduction control parameters (power attenuation coefficient of 0.443, transition time of 3 seconds) and external disturbance abnormal audible and visual alarm data. After the system completes command packet verification and channel stripping, the joint servo driver receives the downlink hardware drive command stream and linearly reduces the output power from the rated power to 0.443 times the rated power within 3 seconds, with the power change slope being... Double the rated power per second. At the same moment the power begins to decrease, after receiving the downlink interactive warning data stream, the voice synthesis chip announces "External impact detected, system has activated protection," the LED indicator operates in a high-frequency red flashing mode, and the buzzer sounds at the alarm frequency, clearly informing the wearer that the current abnormality is caused by an external impact and that the system has actively activated protection measures.

[0105] In the third embodiment, assuming the abnormal danger level is Level 3 (severe danger), the multimodal intervention execution command package generated in step five includes joint easing and mechanical locking control parameters. After the system completes command package verification and channel stripping, the joint servo driver first stops the assist output, and then smoothly reduces the joint angular velocity to zero within 1 to 2 seconds using a preset deceleration ramp to prevent the wearer from falling forward due to inertia due to sudden joint locking. Once the joint angular velocity drops to zero, the mechanical locking mechanism is triggered to fix the joint position. At the same moment the joint begins to decelerate, the audio-visual interaction module activates the highest level of audio-visual alarm, the voice synthesis chip cyclically broadcasts the root cause of the abnormality and handling suggestions, the LED indicator flashes red at the highest frequency, and the buzzer continuously alarms until the system receives a manual reset signal.

[0106] In summary, the exoskeleton graded safety detection method with anomaly backtracking according to the embodiments of this application is explained. It achieves accurate and structured judgment of anomalies through multi-feature constraint consistency evaluation, achieves automatic source tracing and interpretable output of anomaly root causes through pattern mapping of violation sets, and achieves fine-grained control of safety handling intensity through gradient gradual intervention driven by violation degree. This solves the technical problems in the prior art, such as high false positive and false negative rates due to single anomaly judgment dimension, lack of targeted handling due to inability to locate the root cause after anomaly occurs, and direct shutdown leading to wearer imbalance and falls. Ultimately, it achieves a synergistic improvement of the exoskeleton safety detection system in four dimensions: judgment accuracy, source tracing capability, handling safety, and working condition adaptability.

[0107] Furthermore, a graded safety detection device for exoskeletons with anomaly backtracking is also provided.

[0108] Figure 5 This is a block diagram of a graded safety detection device for exoskeletons with anomaly backtracking according to an embodiment of this application. Figure 5 As shown, the exoskeleton graded safety detection device 400 with anomaly backtracking according to an embodiment of this application includes: a feature normalization module 410, used to perform motion pattern classification and feature normalization extraction on the collected multi-source raw data stream of the exoskeleton to obtain a set of working condition constraint parameters and a normalized feature vector of the exoskeleton; a constraint evaluation module 420, used to perform constraint consistency evaluation on the normalized feature vector of the exoskeleton based on the set of working condition constraint parameters to obtain a set of constraint violation indexes and a set of constraint out-of-bounds values; an anomaly tracing module 430, used to perform structural matching tracing on the set of constraint violation indexes based on preset violation pattern mapping rules to obtain the anomaly tracing root cause category; risk classification; The first module 440 is used to quantify the set of constraint violation values ​​and match the threshold range to obtain the constraint violation degree value and the abnormal danger level. The second module 450 is used to select the corresponding gradient intervention strategy according to the abnormal danger level, use the constraint violation degree value as the intensity adjustment weight, and retrieve the corresponding obstacle removal warning information according to the abnormal root cause category. The three are encapsulated and integrated in the same frequency to obtain a multimodal intervention execution instruction package. The third module 460 is used to parse the multimodal intervention execution instruction package and distribute it for execution to drive the exoskeleton joint actuator to implement gradual torque control and synchronously drive the audio-visual interaction module to output the root cause warning broadcast.

[0109] The implementation method of the device involved in this embodiment is the same as that of the method described above, and will not be repeated here.

[0110] According to embodiments of this application, this application also provides an electronic device and a readable storage medium.

[0111] Figure 6 A schematic block diagram of an example electronic device that can be used to implement embodiments of this application is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the application described and / or claimed herein.

[0112] like Figure 6As shown, device 500 includes a computing unit 501, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 502 or a computer program loaded from storage unit 508 into random access memory (RAM) 503. RAM 503 may also store various programs and data required for the operation of device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via bus 504. Input / output (I / O) interface 505 is also connected to bus 504.

[0113] Multiple components in device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0114] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as the exoskeleton graded safety detection method with anomaly backtracking. For example, in some embodiments, the exoskeleton graded safety detection method with anomaly backtracking can be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of the exoskeleton graded safety detection method with anomaly backtracking described above can be performed. Alternatively, in other embodiments, the computing unit 501 may be configured, by any other suitable means (e.g., by means of firmware), to perform an exoskeleton hierarchical safety detection method with anomaly backtracking.

[0115] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0116] The program code used to implement the methods of this application may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0117] In the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on 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 fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0118] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0119] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0120] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0121] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0122] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of those features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0123] 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.

Claims

1. A graded safety detection method for exoskeletons with anomaly backtracking, characterized in that, include: Step 1: Perform motion pattern classification and feature normalization extraction on the collected exoskeleton multi-source raw data stream to obtain the set of working condition constraint parameters and the exoskeleton normalized feature vector; Step 2: Based on the set of working condition constraint parameters, perform constraint consistency evaluation on the normalized feature vector of the exoskeleton to obtain the constraint violation index set and the constraint out-of-bounds value set. Step 3: Based on the preset violation pattern mapping rules, perform structural matching and source tracing on the constraint violation index set to obtain the anomaly root cause category; Step 4: Quantify the set of constraint violation values ​​and match them with threshold intervals to obtain the constraint violation degree value and the anomaly risk level. Step 5: Select the corresponding gradient intervention strategy based on the abnormal danger level, use the constraint violation value as the intensity adjustment weight, and retrieve the corresponding obstacle removal warning information based on the abnormal root cause category. Encapsulate and integrate the three in the same frequency to obtain a multimodal intervention execution instruction package. Step six: Parse and distribute the multimodal intervention execution command package to drive the exoskeleton joint actuator to implement gradual torque regulation and simultaneously drive the audio-visual interaction module to output root cause warning broadcast.

2. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, The exoskeleton's multi-source raw data stream includes joint angle signals, joint angular velocity signals, joint angular acceleration signals, gait phase signals, trunk pitch angle signals, trunk roll angle signals, wear coupling index, dynamic error judgment domain, and sensor communication quality signals.

3. The exoskeleton graded safety detection method with anomaly backtracking according to claim 2, characterized in that, Step one includes: Step 1.1: In response to the trigger signal of the exoskeleton closed-loop feedback control cycle, the multi-source raw data stream of the exoskeleton is decoupled and distributed to obtain the raw motion posture data and the raw feature deviation and state data. Step 1.2: Based on the classification rules, the raw motion posture data is subjected to working condition threshold interval determination and parameter library addressing mapping to obtain the working condition constraint parameter set; Step 1.3: Extract error scalars and perform dimensionless normalization on the feature deviation and original state data to obtain the normalized feature vector of the exoskeleton.

4. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, Step two includes: Step 2.1: Inject the tolerance threshold and sensitivity coefficient from the set of working condition constraint parameters into the preset basic human-machine physical constraint model to obtain a multidimensional instantiated constraint function cluster; Step 2.2: Input the normalized feature vector of the exoskeleton as the independent variable into the multidimensional instantiation constraint function cluster and perform parallel residual analysis on a class-by-class basis to obtain the constraint consistency evaluation quantization matrix. Step 2.3: Based on the zero-value threshold judgment criterion, the constraint consistency evaluation quantization matrix is ​​subjected to out-of-bounds logic screening and numerical index separation. The out-of-bounds dimension numbers are aggregated to obtain the constraint violation index set, and the out-of-bounds magnitude values ​​are packaged to obtain the constraint out-of-bounds value set.

5. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, Step three includes: performing discrete subset structure comparison and anomaly label extraction on the constraint violation index set through the source hash mapping table to obtain the anomaly source root cause category.

6. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, Step four includes: Step 4.1: Perform cumulative calculation of all positive out-of-bounds values ​​in the constraint out-of-bounds value set and quantitative evaluation of the violation degree to obtain the constraint violation degree value; Step 4.2: Based on the preset segmented step threshold, the constraint violation value is range-defined and level-matched to obtain the abnormal risk classification level.

7. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, Step five includes: Step 5.1: Based on the benchmark strategy lookup table, perform macro-intervention strategy matching on the abnormal risk level to obtain the benchmark strategy, and after converting the constraint violation degree value into the control strength adjustment weight, dynamically modulate it with the benchmark strategy to obtain the hardware physical intervention control logic. Step 5.2: Based on the expert troubleshooting database, use the anomaly root cause category as the primary key index to perform retrieval mapping to obtain specific object interaction warning data; Step 5.3: Perform synchronous stamping and protocol encapsulation on the hardware physical intervention control logic and the interactive warning data of specific objects to obtain a multimodal intervention execution instruction package.

8. The exoskeleton graded safety detection method with anomaly backtracking according to claim 1, characterized in that, Step six includes: Step 6.1: Based on the preset frame protocol, the multimodal intervention execution instruction packet is verified, demultiplexed, and channel stripped, wherein the torque control frame is reconstructed into a downlink hardware drive instruction stream, and the interactive warning frame is reconstructed into a downlink interactive warning data stream; Step 6.2: After performing a gradual smoothing filter on the downlink hardware driver command stream, inject it into the joint servo driver for execution, and simultaneously inject the downlink interactive warning data stream into the audio-visual module to execute root cause broadcasting.

9. The exoskeleton graded safety detection method with anomaly backtracking according to claim 6, characterized in that, Step 4.1 includes: Based on the constraint dimension attribution identifier carried by each element in the constraint out-of-bounds value set, the constraint out-of-bounds value set is reconstructed in ascending order of dimension numbering and concatenated into column vectors to obtain an ordered vector of out-of-bounds values. Using the combination of dimension numbers corresponding to each component in the out-of-bounds ordered vector as two-dimensional index pairs, submatrix slicing is performed in the preset global coupling amplification prior matrix to obtain the activation-state coupling amplification weight submatrix. Based on the activated state coupling amplification weight submatrix, a quadratic coupling operation is performed on the ordered vector of out-of-bounds values ​​to obtain the constraint violation degree value.

10. A graded safety detection device for exoskeletons with anomaly backtracking, used to execute the graded safety detection method for exoskeletons with anomaly backtracking as described in any one of claims 1-9, characterized in that, include: The feature normalization module is used to classify motion patterns and extract features from the acquired multi-source raw data stream of the exoskeleton, so as to obtain the set of working condition constraint parameters and the normalized feature vector of the exoskeleton. The constraint evaluation module is used to evaluate the constraint consistency of the exoskeleton normalized feature vector based on the set of working condition constraint parameters to obtain the constraint violation index set and the constraint out-of-bounds value set. The anomaly tracing module is used to perform structural matching and tracing of the constraint violation index set based on preset violation pattern mapping rules to obtain the anomaly root cause category. The risk classification module is used to quantify the set of constraint violation values ​​and match them with threshold intervals to obtain the constraint violation degree value and the abnormal risk level. The strategy encapsulation module is used to select the corresponding gradient intervention strategy based on the level of abnormal danger, use the constraint violation value as the intensity adjustment weight, and retrieve the corresponding troubleshooting warning information based on the abnormal root cause category. The three are encapsulated and integrated in the same frequency to obtain a multimodal intervention execution instruction package. The instruction execution module is used to parse and distribute multimodal intervention instruction packages for execution to drive the exoskeleton joint actuators to implement gradual torque control and simultaneously drive the audio-visual interaction module to output root cause warning broadcasts.

Images