Parallel robot performance evaluation method and system

Abstract

The application relates to the technical field of robot performance evaluation, and discloses a parallel robot performance evaluation method and system, which comprises the following steps: S1, collecting operation data of the parallel robot, preprocessing the operation data, obtaining operation state data, generating a working condition segment, constructing a time window sequence, and establishing a state point set according to the operation state data; S2, extracting a topological feature vector pair for each state point set; S3, calculating a structure drift amount according to the working condition segment, and generating a time window level performance index; S4, parallelly calculating a structure model performance index of each time window, executing structure model excellent determination, and outputting a parallel robot performance evaluation result based on the time window level performance index and the structure model excellent determination result. The method can output stable and comparable performance evaluation results in the actual task operation process of the parallel robot, and can reduce false positives and false negatives caused by speed changes, control state switching and short-time disturbances.

Description

Technical Field

[0001] This invention relates to the field of robot performance evaluation technology, specifically to a method and system for evaluating the performance of parallel robots. Background Technology

[0002] Parallel robots possess characteristics such as high rigidity, high load-bearing capacity, and high dynamic response, and are widely used in scenarios such as high-speed sorting, precision assembly, and machining. These scenarios have high requirements for stability and consistency. On-site engineering often requires objective evaluation of performance status during task execution to promptly detect performance degradation, avoid quality fluctuations, and provide a basis for maintenance and calibration.

[0003] Existing evaluation methods typically focus on two types of information: one is configurational performance indicators based on structural models, such as evaluating the workspace using velocity mapping sensitivity, singularity proximity, or stiffness-related indicators; the other is single-channel amplitude characteristics based on operational data, such as tracking error, drive current, or threshold determination of vibration amplitude. Both methods have their value in engineering applications, but they are prone to limitations under complex operating conditions: structural model indicators tend to focus on theoretical configurational capabilities and cannot adequately cover factors such as assembly errors, friction variations, clearance variations, external load disturbances, and changes in measurement quality; single-channel amplitude characteristics are easily affected by operating condition switching, velocity range changes, and short-term disturbances, leading to false alarms or missed alarms.

[0004] Furthermore, a common pain point exists on engineering sites: structural model indicators show excellent performance, but operational data exhibits abnormalities, or operational data remains stable while structural model indicators suggest risks. Common practices often employ weighted fusion or voting methods to force a single conclusion, lacking an actionable mechanism for handling conflicting scenarios. This results in insufficient interpretability and reproducibility of assessment results, making it difficult to support stable engineering decisions. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for evaluating the performance of a parallel robot, comprising:

[0007] Step S1: Collect and preprocess the operation data of the parallel robot to obtain the operation status data. Generate working condition segments based on the operation status data. Construct a time window sequence within the working condition segments. Establish a set of state points within each time window based on the operation status data.

[0008] Step S2: Extract topological feature vector pairs for each set of state points;

[0009] Step S3: Calculate the structural drift based on the topological feature vectors for each working condition segment, and generate time window-level performance indicators based on the structural drift.

[0010] Step S4: Calculate the structural model performance index for each time window in parallel, perform structural model excellence determination according to the working condition segment, and output the parallel robot performance evaluation result based on the time window level performance index and the structural model excellence determination result.

[0011] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the operating status data includes joint position data, joint velocity data, drive current data, end-effector pose error data, vibration acceleration data, control mode marker data, and trajectory command data.

[0012] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the step of generating working condition segments based on operating status data includes:

[0013] Combine control mode labeling data and joint velocity data to generate operating condition labels;

[0014] Traverse the running status data along the time axis, and generate new segment boundaries when the operating condition label changes;

[0015] A continuous time interval between adjacent segment boundaries is recorded as a working condition segment, and the start time, end time and corresponding working condition label are recorded for each working condition segment.

[0016] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the step of extracting topological feature vector pairs for each set of state points includes:

[0017] Arrange the state points in the state point set in chronological order;

[0018] With each state point as a node, for any node i, establish a directed edge from node i to node j. Each node only establishes directed edges with the previous H nodes in the future. Define the edge weight for each directed edge to obtain a directed state transition graph.

[0019] For any pair of nodes (i, j), calculate the shortest directed path length from node i to node j, and obtain the shortest directed path distance matrix;

[0020] Based on the shortest directed path distance matrix, calculate the symmetric distance for any pair of nodes (i, j) to obtain the time-constrained symmetric distance matrix;

[0021] A distance threshold sequence is set based on the time-constrained symmetric distance matrix. A complex sequence is constructed based on the time-constrained symmetric distance matrix and the distance threshold sequence, and the persistence cohomology is calculated to obtain 0-persistence long-term information and 1-persistence long-term information.

[0022] The 0-dimensional persistent information is converted into a 0-dimensional topological feature vector, and the 1-dimensional persistent information is converted into a 1-dimensional topological feature vector to obtain a topological feature vector pair.

[0023] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the step of calculating the structural drift based on the topological feature vector for each work segment includes:

[0024] According to the working condition label to which the working condition segment belongs, obtain the 0-dimensional reference vector and the 1-dimensional reference vector;

[0025] Calculate the 0-dimensional structural drift based on the 0-dimensional topological feature vector and the 0-dimensional reference vector, and calculate the 1-dimensional structural drift based on the 1-dimensional topological feature vector and the 1-dimensional reference vector;

[0026] The maximum value between the 0-dimensional structural drift and the 1-dimensional structural drift is taken as the structural drift value of the time window.

[0027] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the step of generating time window-level performance indicators based on structural drift includes:

[0028] For the working condition label corresponding to the time window, set the drift threshold corresponding to the working condition label, and subtract the structural drift amount from the drift threshold to obtain the structural margin.

[0029] When the structural margin is greater than or equal to 0, the time window-level performance index is 1, indicating that the structural drift in the current time window has not exceeded the allowable upper limit.

[0030] When the structural margin is less than 0, the time window level performance index is 0, indicating that the structural drift in the current time window exceeds the allowable upper limit.

[0031] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the calculation of the structural model performance index for each time window and the determination of the structural model's quality according to the working condition segment include:

[0032] For each time window, the joint position data corresponding to the center moment of the time window is taken as the configuration data of that time window;

[0033] Based on the kinematic model of the parallel robot, the velocity Jacobian matrix is ​​calculated under configuration data, and the condition number of the velocity Jacobian matrix is ​​used as the performance index of the structural model.

[0034] For the working condition label corresponding to the time window, set the model good or bad threshold corresponding to the working condition label. When the performance index value of the structural model is not higher than the model good or bad threshold, the model is judged to be good or bad for that time window.

[0035] When the performance index of the structural model is higher than the threshold for good or bad, the model is deemed not good or bad for that time window.

[0036] The percentage of time windows in a statistical working condition segment where the model is not excellent is determined. When this percentage exceeds a preset threshold, the structural model in that working condition segment is judged as not excellent.

[0037] As a preferred embodiment of the parallel robot performance evaluation method described in this invention, the output of the parallel robot performance evaluation result based on the time window-level performance index and the structural model excellence judgment result includes:

[0038] For each working condition segment, collect all time window records that fall within the time range of that working condition segment and have the same working condition label. The time window records include time window-level performance indicators and the structural model goodness judgment results within the time window.

[0039] Within this operating condition segment, the length of consecutive occurrences of a time window-level performance index of 0 is counted, and the maximum value of the consecutive lengths is taken as the over-limit duration of this operating condition segment.

[0040] When the duration of the over-limit reaches the preset length threshold, the performance index of the working condition segment is determined to be that the operating structure has exceeded the limit.

[0041] Within this operating condition segment, if the running structure does not exceed the limit, but the structural model in the operating condition segment is judged as not excellent, the performance evaluation result of the operating condition segment is judged as unreliable.

[0042] Except for cases where the operational structure exceeds limits or the assessment is unreliable, the performance indicators of the operating condition segment are normal for the operational structure.

[0043] After obtaining the performance evaluation results of the working condition segments, the performance evaluation results of all working condition segments within the same task cycle are compiled as the performance evaluation results of the parallel robot.

[0044] A parallel robot performance evaluation system employing any of the methods described in this invention, wherein: a data module collects and preprocesses the operation data of the parallel robot to obtain operation status data, generates working condition segments based on the operation status data, constructs a time window sequence within the working condition segments, and establishes a set of state points within each time window based on the operation status data;

[0045] The extraction module extracts topological feature vector pairs for each set of state points;

[0046] The drift module calculates the structural drift amount based on the topological feature vector for each working condition segment, and generates time window-level performance indicators based on the structural drift amount.

[0047] The output module performs parallel calculations of the structural model performance indicators for each time window, performs structural model excellence determination according to the working condition segment, and outputs the performance evaluation results of the parallel robot based on the time window-level performance indicators and the structural model excellence determination results.

[0048] The beneficial effects of this invention are as follows: The method of this invention can output stable and comparable performance evaluation results during the actual operation of parallel robots, reducing false alarms and missed alarms caused by speed changes, control state switching, and short-term disturbances; at the same time, the conclusions on the operational status data side and the conclusions on the structural model side form a mutual verification relationship. When the two are inconsistent, the evaluation is output as unreliable, avoiding the erroneous conclusions caused by the forced weighted fusion of conventional methods, thereby improving the reliability of the evaluation and the engineering feasibility; finally, the evaluation results correspond to the operating range under the specific working condition label, which facilitates the rapid location of risk stages and working conditions of concern, and improves the field applicability and interpretability. Attached Figure Description

[0049] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is an overall flowchart of a parallel robot performance evaluation method provided in Embodiment 1 of the present invention. Detailed Implementation

[0051] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0052] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for evaluating the performance of a parallel robot is provided, comprising:

[0053] Step S1: Collect and preprocess the operation data of the parallel robot to obtain the operation status data. Generate working condition segments based on the operation status data. Construct a time window sequence within the working condition segments. Establish a set of state points within each time window based on the operation status data.

[0054] Using the operational data of parallel robots performing actual tasks as the basic input for performance evaluation, the operational data is preprocessed to obtain operational status data. The operational data comes from real-time sampling records of the controller, actuators, and sensors. The acquisition end writes a unified timestamp to each sampling point, making different data channels comparable on the same timeline.

[0055] To avoid the impact of differences in sampling frequency and sampling time on data comparability, time alignment processing is performed on the operational data. Time alignment processing constructs a regular time axis with a unified sampling period. The earliest timestamp in the operational data is taken as the starting alignment time. Then, using the unified sampling period as the time interval, subsequent alignment times are generated sequentially at fixed intervals starting from the starting alignment time. Generation stops when a generated alignment time exceeds the latest timestamp in the operational data. The resulting sequence of alignment times arranged at fixed intervals constitutes the regular time axis. The operational data is mapped onto the regular time axis, and resampling and interpolation are used to ensure the continuity and integrity of the operational data.

[0056] To ensure consistency in amplitude scale across data collected from different batches and time periods, amplitude normalization is further performed on the time-aligned operational data to obtain operational status data. The statistics for amplitude normalization are derived from the baseline operational status dataset. The baseline operational status dataset refers to the collection of operational status data collected when the parallel robot performs its tasks after passing acceptance testing.

[0057] The operational status data includes joint position data, joint velocity data, drive current data, end-effector pose error data, vibration acceleration data, control mode marker data, and trajectory command data. Joint position data represents the configuration of each drive joint; joint velocity data represents the motion state and supports the division of working conditions; drive current data reflects load changes and drive response; vibration acceleration data reflects the dynamic response of the structure; control mode marker data reflects changes in control state; trajectory command data represents the desired trajectory and stage information; and end-effector pose error data represents the deviation between the actual pose and the desired pose of the parallel robot's end effector during task execution. The pose includes both position and orientation. The end-effector pose error data uses the desired trajectory as a reference, calculates the actual end-effector pose using joint position data combined with a known kinematic model of the parallel robot, and then subtracts the two to obtain the end-effector pose error data.

[0058] Furthermore, operational status segments are generated based on the operational status data. During task execution, the control mode and motion speed of the parallel robot switch according to the task stage. Switching control modes alters the control loop structure, target quantities, and constraints, resulting in significant differences in the statistical distribution and coupling relationships of operational status data such as joint drive current, end-effector pose error, and vibration acceleration. Within the same control mode, changes in speed level further alter inertial force, friction effects, and structural vibration response, thus affecting the dynamic characteristics of the operational status data. Based on these mechanisms, control mode marking and joint speed are used as the primary criteria for operational status segmentation. This decomposes the operational status data into multiple internally consistent time intervals, ensuring comparability when comparing structures within similar operational conditions and avoiding misjudging normal operational status transitions as performance degradation.

[0059] Specifically, control mode marker data and joint speed data are acquired. The control mode marker data is the current output mode number or mode status code of the controller, which is directly used as part of the operating condition label. The joint speed data is used to generate speed levels. The speed level is obtained by the magnitude of the joint speed and the corresponding preset threshold range. For example, the joint speed magnitude is divided into three categories: low speed, medium speed, and high speed, each corresponding to a speed level number.

[0060] During the traversal along the time axis, the system calculates the current operating condition label at each sampling moment and compares it with the operating condition label at the previous sampling moment. When the control mode label changes, the current operating condition label will inevitably change, and the system directly generates the segment boundary at that moment. When the control mode label does not change but the joint speed crosses the speed level threshold, the operating condition label also changes, and the system also generates the segment boundary at that moment. After completing the traversal, the continuous time interval between adjacent boundaries is defined as an operating condition segment, and the start time, end time, and operating condition label are written for this segment. Each operating condition segment obtained in this way simultaneously satisfies "consistent control mode and consistent speed level," which is used to construct time windows and establish state point sets within the segment, avoiding treating normal data differences caused by mode switching or speed segment switching as performance degradation.

[0061] For example, at the beginning of the task, the controller is in position tracking mode, with the control mode number recorded as 1. The joint speed is in the low-speed range, with the speed level number recorded as A. The condition label for this time period is (1, A). When the robot enters the rapid movement phase, the control mode number is 1, but the joint speed crosses the threshold and enters the medium-speed range, the speed level number changes to B. At this time, the condition label becomes (1, B), and the system generates a new segment boundary at the moment the threshold is crossed. Subsequently, the robot enters the fine alignment phase, and the controller switches to force-position hybrid mode, with the mode number recorded as 2. Even if the joint speed is still in the medium-speed range, the condition label will change to (2, B), and the system generates a new segment boundary at the moment of mode switching. Finally, three condition segments are obtained, corresponding to (1, A), (1, B), and (2, B) respectively. Each segment records the corresponding start and end times and label, providing a stable and reproducible segmentation basis for subsequent comparisons of the same condition.

[0062] Control mode labels directly reflect the controller's operating state and control objectives, enabling the differentiation of data generation processes under different control mechanisms at the source. Joint velocity reflects motion intensity, determining the levels of inertial terms, velocity-related damping, friction, and vibration excitation. Together, these constitute operating condition labels, which can segment operational data into intervals with consistent control mechanisms and motion intensity. This makes the data coupling structure within the same operating condition segment more stable, thereby improving the stability and interpretability of subsequent structural drift assessments.

[0063] Furthermore, after obtaining the operating condition segments, a time window sequence is constructed within each segment. The time window is used to divide the continuous operating state data within the operating condition segment into several small intervals of equal length, so that the subsequent establishment of state point sets and feature extraction can be carried out under a unified data scale and time scale.

[0064] Specifically, for any given work condition segment, time windows are sequentially extracted starting from the start time of that segment, according to a preset window length and window step size: the start time of the first time window is the start time of the work condition segment, and the end time is the start time plus the window length; the start time of the next time window is the start time of the previous time window plus the window step size, and the end time is also the start time plus the window length. This process continues along time according to the above rules until the end time of the time window exceeds the end time of the work condition segment. Each time window records its start time, end time, and associated work condition label.

[0065] Within each time window, a set of status points is established based on the operational status data. This set is used to convert the operational status data within that time window into a set of data that can be used for subsequent structural analysis.

[0066] For any time window, first, within the start time and end time corresponding to the time window, intercept the time-aligned operating state data to obtain the synchronous sampling point sequence within the time window. The synchronous sampling point sequence contains multiple sampling moments, and each sampling moment simultaneously has joint position data, joint velocity data, drive current data, end pose error data, and vibration acceleration data. Subsequently, for each sampling moment within the time window, construct a state point in a fixed order. The fixed order does not have to be an unchanging order, and the order between the data can be arranged arbitrarily, as long as an order is kept unchanged during construction and subsequent analysis.

[0067] Combine the state points of all sampling moments within the time window into the state point set of the time window. Each state point in the state point set corresponds to a sampling moment, and the number of state points is equal to the number of valid sampling points within the time window. Through the above processing, the multi-channel operating state data originally arranged by time within the time window is converted into the form of a state point set. The state point set can reflect the coupling relationship and change pattern between the data geometrically, providing a unified input for subsequent feature extraction from the structural perspective and drift evaluation.

[0068] Step S2: Extract topological feature vector pairs for each state point set.

[0069] After obtaining the state point set of the time window, extract topological feature vectors from the state point set to characterize the structural features of the operating state data and convert the structural features into a fixed dimension for subsequent drift calculation. The state point set consists of multiple state points, each state point corresponds to a sampling moment, and there is a time sequence relationship among the state points within the state point set.

[0070] First, arrange the state points in the state point set in chronological order. The purpose of arranging in chronological order is to retain the state evolution direction, so that the subsequent constructed structural features not only reflect the geometric distribution of the state points but also reflect the way the state points transfer over time.

[0071] Construct a directed state transition graph with each state point as a node. For any node i, only establish directed edges with the first H nodes in the future, that is, establish a directed edge from node i to node j, satisfying i < j and j - i ≤ H.

[0072] It should be noted that since each state point corresponds to a unique sampling moment and the state points are arranged in ascending order of sampling moments, when i < j, the sampling moment corresponding to the i-th state point is earlier than the sampling moment corresponding to the j-th state point, and there is a definite chronological relationship between the two on the time axis. Therefore, i and j can be compared, and i < j is used to ensure that the directed edges are only established forward in time. The threshold H for the number of forward connection steps is obtained by dividing the maximum forward time span by the unified sampling period. If it cannot be divided evenly, it is rounded down. The maximum forward time span is set as the time constant when the joint speed reaches the rated speed (the upper limit of the speed for long-term stable operation of the joint, obtained from the instruction manual of the parallel robot) within this operating condition segment, so as to ensure that a directed edge covers an achievable state transition within a short period of time and at the same time avoid forcibly connecting long-distance points in different task stages or different speed segments.

[0073] Define the edge weight for each directed edge. The edge weight is used to quantify the magnitude of the state change from node i to node j. The edge weight is calculated using the two-norm distance:

[0074] ;

[0075] where represents the edge weight from node i to node j; represents the state point of node j; represents the state point of node i; represents the two-norm. Through the above construction, a directed state transition graph is obtained. The constraint that the directed edges only point to future nodes enables the graph structure to carry time direction information, thus injecting the dynamic transfer characteristics within the time window into the subsequent structure analysis.

[0076] Then, calculate the shortest directed path length from node i to node j for any node pair (i, j), and combine the shortest directed path lengths of all node pairs into a shortest directed path distance matrix. The shortest directed path length is defined as the minimum value of the sum of the path edge weights among all directed paths from node i along the directed edges to reach node j. A directed path refers to an edge sequence in the directed state transition graph that starts from node i and passes through several nodes in turn along the direction of each directed edge and finally reaches node j.

[0077] When there is no directed path from i to j, the shortest directed path length is recorded as infinity. The shortest directed path length extends the "direct difference between state points" to the "difference of the transfer chains allowed along the time sequence", thus being more sensitive to such operating state changes as "transfer detour, transfer slowdown, and transfer path change".

[0078] It should be noted that when establishing directed edges for the first H adjacent nodes, direct edges only describe a single change in a short time. Calculating the shortest directed path length for any node pair involves calculating the cumulative cost of multiple consecutive transitions, used to characterize the overall transition structure from any state at any given time to any subsequent state within a time window. This reflects the "gradual deviation, detour, and stagnation" phenomena that occur in parallel robots during operation: for example, under the same working condition, due to increased friction or changes in mechanism clearance, the state does not change abruptly, but gradually deviates through multiple small changes. The shortest directed path length increases with this cumulative deviation, thus revealing a decline in performance margin earlier.

[0079] To ensure that the subsequent complex sequence construction satisfies the symmetric distance requirement, the symmetric distance is calculated based on the shortest directed path distance matrix to obtain the time-constrained symmetric distance matrix. For any pair of nodes (i, j), the symmetric distance is defined as:

[0080] ;

[0081] in, Let represent the symmetric distance between (i, j); min indicates taking the minimum value; This represents the length of the short directed path from node i to node j; This represents the length of the short directed path from node j to node i.

[0082] A time-constrained symmetric distance matrix is ​​formed by the symmetric distances of all node pairs. This symmetry-enhancing process preserves the time constraint information while ensuring the symmetry of the distances, facilitating the subsequent establishment of connection relationships and construction of complex sequences based on distance thresholds.

[0083] Next, a distance threshold sequence is set based on the time-constrained symmetric distance matrix. A complex sequence is constructed based on the time-constrained symmetric distance matrix and the distance threshold sequence, and sustained cohomology is calculated to obtain 0-sustainability information and 1-sustainability information. The distance threshold sequence is arranged in ascending order, covering the effective distance range in the time-constrained symmetric distance matrix. For each distance threshold, if the symmetric distance between node pair (i, j) is less than or equal to the time threshold, node i and node j are considered connected.

[0084] Using the Lipses complex technique, based on the connection relationship, any pairwise connected nodes are elevated to a high-dimensional simplex, thus forming a complex at the distance threshold; as the distance threshold increases, a complex sequence is obtained.

[0085] A reduction-order elimination method based on boundary operators is used to calculate the continuous homology of the complex sequence. A boundary matrix is ​​constructed for the complex sequence and matrix elimination is performed to obtain the generation distance threshold and disappearance distance threshold of each topological feature, thereby obtaining 0-persistence longevity information and 1-persistence longevity information.

[0086] Zero-duration information corresponds to changes in connected components, which represent a group of nodes that can be connected to each other through connections at the current distance threshold. As the distance threshold increases, the originally separate connected components will gradually merge. Mapped to the operating state of a parallel robot, zero-duration information reflects the change in the "connectivity" between state points within a time window with the distance scale: if a large distance threshold is required for a large number of mergers to occur, it indicates that the differences between state points are greater, and the state transitions are more discontinuous in the state space.

[0087] The duration information corresponds to changes in the loop structure, which indicates the existence of a closed loop in the connection relationship. When the distance threshold increases, the closed loop will be generated and will be filled and disappear at a larger threshold. Mapped to the operating state of a parallel robot, the duration information reflects whether the state transition within the time window exhibits the loop characteristic of "repeated round trips": when the loop characteristic persists within a wide range of distance thresholds, it indicates that there is a stable closed transition relationship between state points, which is commonly seen in repeated corrections or round trip changes caused by vibration within the same working condition.

[0088] Finally, the 0-dimensional durability information and the 1-dimensional durability information are converted into topological feature vector pairs. The conversion method adopts a fixed-grid vectorization rule: the 0-dimensional durability information is mapped to a fixed-dimensional vector to obtain a 0-dimensional topological feature vector v0, and the 1-dimensional durability information is mapped to a fixed-dimensional vector to obtain a 1-dimensional topological feature vector v1, resulting in a topological feature vector pair (v0, v1). Here, v0 represents the 0-dimensional topological feature vector, and v1 represents the 1-dimensional topological feature vector. The topological feature vector pair serves as the structural representation output of the state point set within the time window, used for subsequent calculation of structural drift.

[0089] Step S3: Calculate the structural drift based on the topological feature vector for each operating condition segment, and generate time window-level performance indicators based on the structural drift.

[0090] After obtaining the topological feature vector pairs corresponding to each time window, a reference vector pair under the same operating condition is acquired. The structural drift is calculated based on the topological feature vector pairs and the reference vector pairs, quantifying the deviation of the running-state coupled structure from the reference state in the current time window. The same operating condition is guaranteed by the operating condition label; identical operating condition labels indicate consistent control state and speed level, reducing interference from operating condition switching on structural comparison and ensuring that the drift primarily reflects changes in the running-state structure rather than changes in the task phase.

[0091] The reference vector pairs are generated from the reference operating state dataset. For each operating condition label, the 0-dimensional and 1-dimensional topological feature reference vectors for all time windows under that label are collected. The mean values ​​of the 0-dimensional and 1-dimensional topological feature reference vectors are obtained by calculating the mean values ​​of each dimension. These mean values ​​are then combined to form the reference vector pairs. The reference vector pairs not only contain the geometric distribution information of the state points in the state space but also information on the typical transition modes of the state over time. In engineering terms, this can be understood as follows: under the same operating condition that has passed acceptance testing, there exists a relatively stable cooperative change pattern among joint position, joint velocity, drive current, end-effector pose error, and vibration acceleration. The reference vector pairs are the numerical characterization of this cooperative change pattern.

[0092] It should be noted that, although there may be slight differences in measurement noise, minor load fluctuations, and trajectory tracking in the baseline operating state dataset, under the same control mode and the same speed level constraint, these slight differences mainly manifest as random fluctuations around a certain stable structural morphology. Taking the mean value dimension by dimension is equivalent to using the common features of a large number of time windows to offset occasional fluctuations, making the obtained baseline vector pair closer to the "typical structural state of a qualified operating state" under this condition, thereby avoiding the bias of a single sample due to noise.

[0093] In parallel robots operating under the same conditions, changes in transmission backlash, friction characteristics, stiffness, drive performance, sensing quality, or external load conditions often do not manifest as a single data point increasing or decreasing. Instead, they frequently exhibit changes in the previously stable relationships between multiple data points. For example, the relationship between current and velocity may become inconsistent, the synchronization between error and vibration may change, and state transitions may involve detours or splits. Since topological feature vector pairs are generated from the state transition structure within a time window, changes in the relationships between these data points directly alter the structural representation of the topological feature vector pairs, thus increasing the difference between the topological feature vector pairs and the reference vector pairs.

[0094] Based on this correspondence, using structural drift as a time window-level judgment criterion can distinguish between actual structural changes and normal noise fluctuations under the same operating conditions. Specifically, for any current time window, given the topological feature vector pair for that time window, and the baseline vector pair labeled under the same operating conditions, the 0-dimensional structural drift and 1-dimensional structural drift are calculated respectively:

[0095] ;

[0096] ;

[0097] Where D0 represents the 0-dimensional structural drift; v0 represents the 0-dimensional topological feature vector; D1 represents the mean of the 0-dimensional topological feature reference vector; D1 represents the 1-dimensional structural drift; v1 represents the 1-dimensional topological feature vector. This represents the mean of the 1-dimensional topological feature reference vector.

[0098] To ensure that the structural drift is sensitive to both 0-dimensional and 1-dimensional structural changes, the maximum value of the 0-dimensional and 1-dimensional structural drift is taken as the structural drift for the time window.

[0099] Furthermore, after obtaining the structural drift amount for each time window, a time window-level performance index is generated, converting the continuous numerical value of structural drift into a window-level result that can be directly used for evaluation and judgment. The structural drift amount reflects the degree of deviation of the running coupled structure from the reference state within the time window. The greater the deviation, the more unstable the transfer relationship between the channels under the same working condition, which is manifested in the change of the correlation structure between the driving load, tracking deviation, and vibration response. This suggests that the operating state of the parallel robot within that time window may have approached or exceeded the acceptable performance boundary.

[0100] The generation of time window-level performance indicators is based on the premise of comparability under the same operating conditions. For each operating condition label, the baseline operating state dataset contains a large number of time windows collected under the condition of acceptance. The structural drift distribution of the baseline window can characterize the "allowable range of structural fluctuations in the qualified operating state". This distribution is used to set a drift threshold for the same operating condition, so that the threshold can adapt to the differences in different operating conditions, thereby avoiding misjudgment caused by using a global threshold to cover all operating conditions.

[0101] Specifically, for each operating condition label, a set of structural drift values ​​with the same label is compiled from the baseline operating state dataset. The high quantile of this set (e.g., 0.99) is calculated as the drift threshold. The drift threshold represents the upper bound of structural drift value under the same operating condition and acceptance-qualified operating state. For the current time window, its structural drift value is read, and the structural drift value is subtracted from the drift threshold to obtain the structural margin. The structural margin is the remaining amount of the current window from the "allowable upper bound for the same operating condition". A non-negative structural margin indicates that it is still within the allowable range, while a negative structural margin indicates that it has exceeded the allowable range.

[0102] Based on structural margin, a time-window-level performance index is generated to represent the time-window-level deliverability determination flag. When the structural margin is greater than or equal to 0, the time-window-level performance index is 1; when the structural margin is less than 0, the time-window-level performance index is 0. The time-window-level performance index and structural margin are used as operational structural evidence output. The time-window-level performance index is used for rapid determination, and the structural margin is used to characterize the margin size and risk level.

[0103] The performance of time window-level metrics in parallel robots can be understood from the deliverability of the operating state. When the time window-level performance metric is 1, the structural drift does not exceed the drift threshold under the same working condition, which means that the state transition structure within this time window is highly consistent with the baseline operating state. Common manifestations include: under the same control mode and the same speed level, the driving current changes with the joint speed in a stable manner; the fluctuation level and fluctuation pattern of the end-effector pose error are within the baseline range; the vibration acceleration does not show a continuous increase or abnormal transfer path within the window; and the system is in a stable operating state that can be maintained.

[0104] When the time window-level performance index is 0, the structural drift exceeds the drift threshold under the same operating condition, indicating that the state transition structure deviates significantly from the reference within that time window. Common manifestations include: abnormal changes in the coupling relationship between the drive current and the end-effector pose error under the same control mode and speed level; continuous amplification or increased jitter in the end-effector pose error; and the correlation between the vibration response and the drive load exhibiting detours or splits, meaning the motion process no longer evolves along the reference transition path. These situations typically correspond to operational structural anomalies caused by factors such as friction changes, clearance changes, structural stiffness changes, decreased sensor measurement quality, or increased external load disturbances. Since this invention only outputs a window-level deliverable determination flag and does not directly provide the fault category, the time window-level performance index is used to indicate that "the performance state under this operating condition has exceeded the structural boundary of the qualified operating state," and to provide data-side evidence for subsequent parallel comparison with the structural model performance index and for unreliable determination.

[0105] Step S4: Calculate the structural model performance index for each time window in parallel, perform structural model excellence determination according to the working condition segment, and output the parallel robot performance evaluation result based on the time window level performance index and the structural model excellence determination result.

[0106] The kinematic performance of parallel robots varies significantly across different configurations, especially in regions approaching singular configurations. The mapping between end-effector velocity and joint velocities becomes sensitive, meaning even small joint perturbations can induce large end-effector responses, and control errors and drive loads are more easily amplified. To introduce comparable structural model evidence into operational evaluation, structural model performance indicators are calculated in parallel while generating time-window-level performance indicators. A structural model quality assessment is then provided at the operational segment scale, forming a comparable evaluation basis with operational structural evidence.

[0107] Specifically, the joint position data corresponding to the center moment of the time window is taken as the configuration data for that time window, representing the main structural state within that time window. Based on the kinematic model parameters of the parallel robot, the velocity mapping matrix is ​​calculated under this configuration data, and the condition number of the velocity mapping matrix is ​​used as the performance index of the structural model. The condition number is used to characterize the sensitivity of the velocity mapping; the larger the condition number, the more unstable the velocity mapping, and the closer the configuration is to the singular region; the smaller the condition number, the more stable the velocity mapping, and the better the kinematic performance in the sense of the structural model. This structural model performance index belongs to a known kinematic analysis method and is used to provide a quantitative representation of the quality of the configuration at the model level.

[0108] To ensure comparability of structural model performance assessments under the same operating conditions, a good / good threshold is established based on operating condition labels. The baseline operating state dataset consists of repeated task operation data under accepted operating conditions, covering the normal fluctuation range under the same operating condition label. For each operating condition label, the set of structural model performance index values ​​corresponding to that operating condition label in the baseline operating state dataset is compiled, and the upper bound value (e.g., 0.95) that covers the vast majority of the baseline windows is selected as the good / good threshold for that operating condition label. Under the same control mode and speed level, normal configuration fluctuations in different task batches can lead to dispersion in structural model performance indicators. Using an upper bound that covers the vast majority of baseline samples as the threshold can accommodate normal fluctuations within the good / good range, avoiding misjudging normal operating condition differences as model degradation.

[0109] At the time window scale, when the structural model performance index of a certain time window is not higher than the model excellence threshold corresponding to its working condition label, the model of that time window is judged to be excellent; when the index is higher than the threshold, the model of that time window is judged to be poor. Subsequently, at the working condition segment scale, a summary judgment is performed. For each working condition segment, the proportion of time windows with poor model performance within the segment is statistically analyzed. When this proportion is not higher than a preset proportion threshold, the model of that working condition segment is judged to be excellent; when this proportion is higher than the preset proportion threshold, the model of that working condition segment is judged to be poor. The reason for using proportion summary instead of single-point judgment is that there are discrete errors between configuration sampling and model calculation. Occasional single-window anomalies may be caused by short-term noise or short-term trajectory changes. Proportion judgment can reflect the overall configuration quality within the segment, which is more in line with the overall evaluation needs of the structural state of the working condition segment.

[0110] Through the aforementioned parallel computation and judgment, this invention obtains structural model evidence in addition to operational structural evidence. Structural model evidence reflects the stability and singularity risk of configuration in a kinematic sense, while operational structural evidence reflects the transfer structural stability of multi-channel operational data. These two types of evidence have different sources and focuses, providing a basis for outputting unreliable assessment results in scenarios of conflicting evidence. This ensures that the assessment conclusion includes both operational factual evidence and structural model comparison evidence.

[0111] Furthermore, after obtaining the performance indicators and structural model excellence criteria at the time window level, performance evaluation results are generated at the operating condition segment scale. Within each operating condition segment, the control mode and speed level remain consistent, and the mechanism for generating operating status data is relatively stable. Therefore, summarizing at this scale can reduce occasional misjudgments caused by short-term disturbances within a single time window, making the evaluation results closer to the actual operating state of that operating condition segment.

[0112] Specifically, for each operating condition segment, all time window records within the segment's time range are collected, ensuring that the operating condition label of the time window record is consistent with that operating condition segment. The time window records contain two types of judgment results: time window-level performance indicators generated by structural drift, and structural model excellence judgment results obtained from structural model performance indicators.

[0113] First, based on time window-level performance indicators, the length of consecutive occurrences of time window-level performance indicators being 0 within a work condition segment is statistically analyzed, and the maximum consecutive length is taken as the over-limit duration length of the work condition segment. When the over-limit duration length reaches a preset length threshold, the performance evaluation result of that work condition segment is directly determined as the running structure exceeding the limit. Since there may be short-term impacts, short-term trajectory changes, or transient anomalies caused by measurement noise during the operation of parallel robots, these transient anomalies will cause a temporary increase in the amount of structural drift, but do not necessarily mean that the performance state of that work condition segment has deteriorated.

[0114] It should be noted that the preset length threshold can be determined by statistical calibration based on the benchmark operating state dataset, ensuring that the threshold is consistent with the sampling period, window length, window step size, and normal fluctuation level under the same operating conditions. In the benchmark operating state dataset, the continuous length of the time window-level performance index exceeding the limit is statistically analyzed for each operating condition label, resulting in a continuous length sample set under that operating condition label; the upper bound (e.g., 0.95) covering the vast majority of benchmark samples in the sample set is selected as the preset length threshold.

[0115] Secondly, when a work condition segment does not trigger a structural overshoot in the operational state, but the structural model in the segment is judged as non-excellent, the performance evaluation result of the work condition segment is deemed unreliable. A non-excellent work condition segment model indicates that the overall configuration corresponding to that work condition segment is in or near the sensitive region of the structural model. In this case, the parallel robot is more amplified by external load disturbances and friction changes. Although the operational structure is not currently exceeding limits, its stability margin is lower, and it is more likely to cross the threshold for slight changes. Directly outputting that the operational structure is normal would mask the fact that the parallel robot's structure is in a highly sensitive state. Therefore, this segment is marked as unreliable to indicate that the conclusion of this work condition segment needs to be reviewed, avoiding the equating of potential risk states with stable and normal states.

[0116] Finally, except for cases where the operating structure exceeds limits and the assessment is unreliable, the performance indicators of the operating condition segment are normal.

[0117] Furthermore, after obtaining the performance evaluation results of the work condition segments, the performance evaluation results of all work condition segments within the same task cycle are compiled as the performance evaluation results of the parallel robot.

[0118] At the same time, the evaluation results synchronously output the start time, end time, operating condition label, and the duration of the time window-level performance index exceeding the limit for each operating condition segment, which is used to locate the task stage and operating condition where the evaluation anomaly occurred.

[0119] Example 2, in an exemplary embodiment, also provides a parallel robot performance evaluation system, including:

[0120] The data module collects and preprocesses the operation data of the parallel robot to obtain operation status data. It generates working condition segments based on the operation status data, constructs a time window sequence within the working condition segments, and establishes a set of state points within each time window based on the operation status data.

[0121] The extraction module extracts topological feature vector pairs for each set of state points.

[0122] The drift module calculates the structural drift amount based on the topological feature vector for each operating condition segment, and generates time window-level performance indicators based on the structural drift amount.

[0123] The output module performs parallel calculations of the structural model performance indicators for each time window, performs structural model excellence determination according to the working condition segment, and outputs the performance evaluation results of the parallel robot based on the time window-level performance indicators and the structural model excellence determination results.

[0124] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0126] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0127] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0128] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for evaluating the performance of a parallel robot, characterized in that, include: Step S1: Collect and preprocess the operation data of the parallel robot to obtain the operation status data. Generate working condition segments based on the operation status data. Construct a time window sequence within the working condition segments. Establish a set of state points within each time window based on the operation status data. Step S2: Extract topological feature vector pairs for each set of state points; The extraction of topological feature vector pairs for each set of state points includes: Arrange the state points in the state point set in chronological order; With each state point as a node, for any node i, establish a directed edge from node i to node j. Each node only establishes directed edges with the previous H nodes in the future. Define the edge weight for each directed edge to obtain a directed state transition graph. For any pair of nodes (i, j), calculate the shortest directed path length from node i to node j, and obtain the shortest directed path distance matrix; Based on the shortest directed path distance matrix, calculate the symmetric distance for any pair of nodes (i, j) to obtain the time-constrained symmetric distance matrix; A distance threshold sequence is set based on the time-constrained symmetric distance matrix. A complex sequence is constructed based on the time-constrained symmetric distance matrix and the distance threshold sequence, and the persistence cohomology is calculated to obtain 0-persistence long-term information and 1-persistence long-term information. Using the vectorization rule of fixed grid, 0-dimensional persistent information is mapped to fixed-dimensional vector to obtain 0-dimensional topological feature vector v0, and 1-dimensional persistent information is mapped to fixed-dimensional vector to obtain 1-dimensional topological feature vector v1, resulting in topological feature vector pair (v0, v1), where v0 represents 0-dimensional topological feature vector and v1 represents 1-dimensional topological feature vector. Step S3: Calculate the structural drift based on the topological feature vectors for each working condition segment, and generate time window-level performance indicators based on the structural drift. The calculation of structural drift based on topological feature vectors for each operating condition segment includes: According to the working condition label to which the working condition segment belongs, obtain the 0-dimensional reference vector and the 1-dimensional reference vector; The zero-dimensional structural drift is calculated based on a zero-dimensional topological feature vector and a zero-dimensional reference vector, and the one-dimensional structural drift is calculated based on a one-dimensional topological feature vector and a one-dimensional reference vector, expressed as follows: ； ； Where D0 represents the 0-dimensional structural drift; v0 represents the 0-dimensional topological feature vector; D1 represents the mean of the 0-dimensional topological feature reference vector; D1 represents the 1-dimensional structural drift; v1 represents the 1-dimensional topological feature vector. This represents the mean of the 1-dimensional topological feature reference vector; The maximum value between the 0-dimensional structural drift and the 1-dimensional structural drift is taken as the structural drift value of the time window; Step S4: Calculate the structural model performance index for each time window in parallel, perform structural model excellence determination according to the working condition segment, and output the parallel robot performance evaluation result based on the time window level performance index and the structural model excellence determination result; The parallel robot performance evaluation results output based on the time window-level performance indicators and structural model excellence determination results include: For each working condition segment, collect all time window records that fall within the time range of that working condition segment and have the same working condition label. The time window records include time window-level performance indicators and the structural model goodness judgment results within the time window. Within this operating condition segment, the length of consecutive occurrences of a time window-level performance index of 0 is counted, and the maximum value of the consecutive lengths is taken as the over-limit duration of this operating condition segment. When the duration of the over-limit reaches the preset length threshold, the performance index of the working condition segment is determined to be that the operating structure has exceeded the limit. Within this operating condition segment, if the running structure does not exceed the limit, but the structural model in the operating condition segment is judged as not excellent, the performance evaluation result of the operating condition segment is judged as unreliable. Except for cases where the operational structure exceeds limits or the assessment is unreliable, the performance indicators of the operating condition segment are normal for the operational structure. After obtaining the performance evaluation results of the working condition segments, the performance evaluation results of all working condition segments within the same task cycle are compiled as the performance evaluation results of the parallel robot.

2. The method for evaluating the performance of a parallel robot as described in claim 1, characterized in that, The operational status data includes joint position data, joint velocity data, drive current data, end-effector pose error data, vibration acceleration data, control mode marker data, and trajectory command data.

3. The method for evaluating the performance of a parallel robot as described in claim 2, characterized in that, The process of generating operating condition segments based on operating status data includes: Combine control mode labeling data and joint velocity data to generate operating condition labels; Traverse the running status data along the time axis, and generate new segment boundaries when the operating condition label changes; A continuous time interval between adjacent segment boundaries is recorded as a working condition segment, and the start time, end time and corresponding working condition label are recorded for each working condition segment.

4. The method for evaluating the performance of a parallel robot as described in claim 3, characterized in that, The generation of time-window level performance metrics based on structural drift includes: For the working condition label corresponding to the time window, set the drift threshold corresponding to the working condition label, and subtract the structural drift amount from the drift threshold to obtain the structural margin. When the structural margin is greater than or equal to 0, the time window-level performance index is 1, indicating that the structural drift in the current time window has not exceeded the allowable upper limit. When the structural margin is less than 0, the time window level performance index is 0, indicating that the structural drift in the current time window exceeds the allowable upper limit.

5. The method for evaluating the performance of a parallel robot as described in claim 4, characterized in that, The calculation of structural model performance indicators for each time window, and the determination of structural model quality based on operating condition segments, includes: For each time window, the joint position data corresponding to the center moment of the time window is taken as the configuration data of that time window; Based on the kinematic model of the parallel robot, the velocity Jacobian matrix is ​​calculated under configuration data, and the condition number of the velocity Jacobian matrix is ​​used as the performance index of the structural model. For the working condition label corresponding to the time window, set the model good or bad threshold corresponding to the working condition label. When the performance index value of the structural model is not higher than the model good or bad threshold, the model is judged to be good or bad for that time window. When the performance index of the structural model is higher than the threshold for good or bad, the model is deemed not good or bad for that time window. The proportion of time windows in the statistical working condition segment where the model is not excellent is calculated. When this proportion is higher than a preset threshold, the structural model in that working condition segment is judged as not excellent.

6. A parallel robot performance evaluation system, applied to the parallel robot performance evaluation method according to any one of claims 1-5, characterized in that, include, The data module collects and preprocesses the operation data of the parallel robot to obtain operation status data. It generates working condition segments based on the operation status data, constructs a time window sequence within the working condition segments, and establishes a set of state points within each time window based on the operation status data. The extraction module extracts topological feature vector pairs for each set of state points; The drift module calculates the structural drift amount based on the topological feature vector for each working condition segment, and generates time window-level performance indicators based on the structural drift amount. The output module performs parallel calculations of the structural model performance indicators for each time window, performs structural model excellence determination according to the working condition segment, and outputs the performance evaluation results of the parallel robot based on the time window-level performance indicators and the structural model excellence determination results.

Images