Search

CN-122021247-A - Industrial robot positioning error prediction method based on stack learning

CN122021247ACN 122021247 ACN122021247 ACN 122021247ACN-122021247-A

Abstract

The invention belongs to the field of advanced manufacturing of industrial robots, and particularly relates to a stack learning-based industrial robot positioning error prediction method. The method comprises the steps of industrial robot space position data generation, error data acquisition, error prediction model construction based on stack learning, error model parameter optimization, industrial robot positioning error prediction and the like. Compared with a single prediction model, the industrial robot positioning error prediction model based on stack learning has the advantages of high prediction precision and good generalization performance, and a plurality of complementary basic models are integrated through a two-layer integrated model structure, so that single model prediction deviation is reduced, and the prediction precision and adaptability of the error prediction model are improved.

Inventors

  • ZHANG PIN
  • CHEN QIZHI
  • MEN WEIWEI
  • BAI LONG
  • LI TIANXIANG

Assignees

  • 中国航空工业集团公司济南特种结构研究所

Dates

Publication Date
20260512
Application Date
20251224

Claims (8)

  1. 1. An industrial robot positioning error prediction method based on stack learning is characterized by comprising the following steps: S1, generating position data in a robot working space by using Latin hypercube space sampling; S2, acquiring positioning error data of the robot to form a positioning error data set; S3, constructing a two-layer positioning error prediction model based on stack learning; S4, optimizing model parameters; s5, determining the prediction precision of the positioning error prediction model, judging whether the prediction precision meets the requirement of error prediction, if so, executing the step S6, otherwise, returning to the step S4 to re-optimize the parameters of the prediction model; S6, carrying out industrial robot positioning error prediction based on the trained positioning error prediction model.
  2. 2. The method for predicting positioning errors of an industrial robot based on stack learning according to claim 1, wherein the robot workspace in S1 is a Cartesian workspace.
  3. 3. The method for predicting positioning errors of an industrial robot based on stack learning of claim 2, wherein the robot spatial position data is generated in a robot workspace using a base coordinate system of the robot as a reference coordinate system.
  4. 4. The method for predicting the positioning error of the industrial robot based on the stacking learning of claim 1, wherein S2 is characterized in that the tail end of the robot is operated to move to all the space positions point by point based on the space position data of the robot, and the tail end position of the robot is measured based on a laser tracker to obtain the positioning error with the theoretical position, so as to form a positioning error data set.
  5. 5. The method for predicting positioning errors of an industrial robot based on stack learning of claim 4, wherein the positioning error dataset is divided to obtain a first training set and a first test set, and the data in the positioning error dataset is scaled to be within a uniform scale by normalizing the data.
  6. 6. The method for predicting the positioning error of the industrial robot based on stack learning according to claim 5, wherein the two-layer positioning error prediction model in S3 comprises a first-layer basic model and a second-layer integrated model, the basic model has a core task of performing preliminary prediction by using a positioning error data set, and the prediction result is used as an intermediate data set required by training the integrated model, so that the prediction of the positioning error is realized through training and optimization of the integrated model.
  7. 7. The method for predicting the positioning error of the industrial robot based on stack learning of claim 6, wherein the super-parameter optimization range based on the basic models in S4 optimizes the super-parameters of each basic model by a grid optimization method to obtain the optimal model prediction performance.
  8. 8. The method for predicting positioning errors of an industrial robot based on stack learning as set forth in claim 7, wherein the model prediction accuracy analysis is performed by using a mean square error, an average relative error and a decision coefficient in S5.

Description

Industrial robot positioning error prediction method based on stack learning Technical Field The invention belongs to the field of advanced manufacturing of industrial robots, and particularly relates to a stack learning-based industrial robot positioning error prediction method. Background Industrial robots are increasingly used in high-end manufacturing applications, but their absolute positioning accuracy is affected by a variety of error factors. These errors include geometric parameter errors (e.g., link dimensional deviations, joint zero offsets) and non-geometric parameter errors (e.g., joint flexibility, temperature deformations, external force loads, etc.). The manufacturing cost of the high-precision robot body is extremely high, and the precision of the elevator robot can be improved on the premise of not remarkably increasing the hardware cost through an error compensation technology. The existing error compensation method is mainly divided into two types of off-line compensation and on-line compensation. An offline compensation method (such as the AR model prediction method described in CN115990886 a) corrects the trajectory in advance before the robot performs the task, but it is difficult to cope with dynamically changing working conditions. The online compensation method (such as the full life cycle compensation method in CN 118596147B) adjusts the motion of the robot in real time in the processing process, but has high requirement on the real-time performance of the system, and the time lag of the measurement system can influence the compensation effect. The main problems existing in the prior art include: (1) The generalization capability of a single prediction model is limited, and the change of error characteristics in the whole life cycle of the robot is difficult to deal with; (2) The existing method has insufficient consideration on the flexibility error of the base and the joint backlash error, especially under the heavy load working condition; (3) The lack of quantitative assessment of the uncertainty of the error prediction makes it difficult to guarantee the compensation reliability. Industrial robot error prediction techniques are evolving towards multi-model fusion and full life cycle adaptation. CN118596147B proposes a position error prediction and compensation method suitable for the full life cycle of a robot, which can fine tune the prediction model by continuously adding new sample data. In addition, the spatial grid method proposed by the invention of CN114161425A and the like predicts errors through interpolation, but faces the problem of dimension disasters in a high-dimensional space. Disclosure of Invention The invention aims to solve the technical problems that an industrial robot is poor in positioning precision, difficult to be suitable for a high-precision manufacturing and processing scene, low in prediction precision, poor in generalization performance and the like in the traditional positioning error prediction method, and provides the high-precision prediction method for the positioning error of the industrial robot. The invention provides an industrial robot positioning error prediction method based on stack learning, which comprises the following steps: S1, generating position data in a robot working space by using Latin hypercube space sampling; S2, acquiring positioning error data of the robot to form a positioning error data set; S3, constructing a two-layer positioning error prediction model based on stack learning; S4, optimizing model parameters; s5, determining the prediction precision of the positioning error prediction model, judging whether the prediction precision meets the requirement of error prediction, if so, executing the step S6, otherwise, returning to the step S4 to re-optimize the parameters of the prediction model; S6, carrying out industrial robot positioning error prediction based on the trained positioning error prediction model. Advantageously, the robot workspace in S1 is a cartesian workspace. Advantageously, the robot spatial position data is generated within the robot workspace with the base coordinate system of the robot as a reference coordinate system. Advantageously, in S2, the robot tip is operated to move point by point to all spatial positions based on the robot spatial position data, and the position of the robot tip is measured based on the laser tracker to obtain a positioning error with the theoretical position, thereby forming a positioning error dataset. Advantageously, the positioning error dataset is partitioned to obtain a first training set and a first test set, and the data in the positioning error dataset is scaled to a uniform scale by normalizing the data. Advantageously, the two-layer positioning error prediction model in S3 includes a first-layer basic model and a second-layer integrated model, where the basic model has a core task of performing preliminary prediction using the positioning error data set, an