Teng Zhang

Teng Zhang

AI + Theory + Robotic = Anything you want!

Posts by Collection

portfolio

publications

A Knowledge-Embedded End-to-End Intelligent Reasoning Method for Processing Quality of Shaft Parts

Published in November 26-28, 2021

The machining quality of a part is one of the most important factors affecting work effectiveness and service time, and it is closely related to multi-stage manufacturing processes (MMPs). State space model (SSM) is a typical method to analyze error propagation in MMPs, which contains the deep laws of error propagation, but the modeling process is complicated and the perception of quality is afterwards. In actual production, it is difficult to realize the pre-reasoning and control of processing quality. To address the above problems, an end-to-end intelligent reasoning method for processing quality with SSM knowledge embedding is proposed. On the one hand, the knowledge embedded in SSM is used for data simulation, and on the other hand, the end-to-end mapping between measured dimensions and processing quality of each process is realized by an Adaptive Network-based Fuzzy Inference System (ANFIS). In this paper, wall thickness difference (WTD) is used to describe the machining quality of shaft parts, and four sections of four processes are studied. SSM was constructed and validated using workshop data, and the average relative error for the six shafts was 5.54%. In the testing phase of the intelligent reasoning model, the maximum RMSE and MAE of the models for the four processes were 4.47 μm and 3.23 μm, respectively, satisfying the WTD prediction requirements.

Recommended citation: T. Zhang, B. Li, H. Sun, S. Zhao, F. Peng, L. Zhou, R. Yan. (2022). A Knowledge-Embedded End-to-End Intelligent Reasoning Method for Processing Quality of Shaft Parts. In: Liu, H., et al. Intelligent Robotics and Applications. ICIRA 2022. Lecture Notes in Computer Science(), vol 13458. Springer, Cham. https://doi.org/10.1007/978-3-031-13841-6_39 https://doi.org/10.1007/978-3-031-13841-6_39

A transfer learning based geometric position-driven machining error prediction method for different working conditions

Published in November 26-28, 2021

Machining error is one of the most important indicators to evaluate the processing quality of thin-walled parts. With the development of Data Science, the data-driven methods have become popular. But the condition that the model can work accurately on a new task is that the feature space and distribution of the data are the same. A sample-based transfer learning method driven by geometric position is utilized to quickly predict the machining errors of thin-walled parts under different working conditions. This method can fully learn the knowledge related to machining errors contained in the data through model training, and can apply this knowledge to accurately and quickly forecast machining errors under new working conditions. In the experimental scenario, this method has outstanding predictive performance. The average determination coefficient of the four groups of target domain experiments reached 0.96, and the average root mean square error is less than 5.32μm. In addition, this method can shorten the machining error acquisition time to 22% of the original, reducing the dependence on time-consuming and expensive measurements greatly

Recommended citation: T. Zhang, H. Sun, L. Zhou, S. Zhao, F. Peng and R. Yan, "A transfer learning based geometric position-driven machining error prediction method for different working conditions," 2021 27th International Conference on Mechatronics and Machine Vision in Practice (M2VIP), Shanghai, China, 2021, pp. 145-150, doi: 10.1109/M2VIP49856.2021.9665105. https://ieeexplore.ieee.org/abstract/document/9665105

A deep transfer regression method based on seed replacement considering balanced domain adaptation

Published in July 17, 2022

In this work, a deep transfer regression method based on seed replacement considering balanced domain adaptation is proposed. On the one hand, the difference of marginal distribution and conditional distribution of different data is considered simultaneously. On the other hand, the domain knowledge is represented and learned through the seed replacement technology based on clustering actively. The performance of the models was compared between two public datasets (the tool wear dataset and the battery capacity dataset) and one private data set (the robot machining errors dataset). The results have indicated that the proposed method is better in prediction accuracy, and has the characteristics of stable performance and weak dependence on pre-training model.

Recommended citation: T. Zhang, H. Sun, F. Peng, S. Zhao, R. Yan, A deep transfer regression method based on seed replacement considering balanced domain adaptation, Eng. Appl. Artif. Intell., 115 (2022) 105238, https://doi.org/10.1016/j.engappai.2022.105238. https://doi.org/10.1016/j.engappai.2022.105238

CME-EPC A coarse-mechanism embedded error prediction and compensation framework for robot multi-condition tasks

Published in October 11, 2023

In this work, a Coarse-Mechanism Embedded Error Prediction and Compensation framework (CME-EPC) for robot multi-condition tasks is proposed. In this framework, error-generating mechanism models for multi-conditions are built for constructing simulation domain datasets to realize the embedding of robot error mechanism models for different working conditions at the data level. On this basis, we propose an AL-based labelling of few-shots and Clustering-guided balanced domain adaptation transfer learning method to achieve accurate robot error prediction under 10% samples, and in addition a pre-compensation of errors based on the prediction results is developed. The CME-EPC framework was validated on four typical tasks in three operating conditions, achieving the best performance relative to other six prediction methods and two compensation methods. After compensation, the average error is reduced by more than 94% at most, which drives high-precision robot applications.

Recommended citation: T. Zhang, F. Peng, X. Tang, R. Yan, R. Deng, CME-EPC: A coarse-mechanism embedded error prediction and compensation framework for robot multi-condition tasks, Robot. Comput.-Integr. Manuf., 86 (2024) 102675, https://doi.org/10.1016/j.rcim.2023.102675. https://doi.org/10.1016/j.rcim.2023.102675

An active semi-supervised transfer learning method for robot pose error prediction and compensation

Published in November 11, 2023

In this work, an active semi-supervised transfer learning method (ASTL) is proposed for accurate and efficient prediction and compensation of robot pose error by integrating the multi-stage greedy sampling (MGS) and the semi-supervised transfer learning (STL). The robot pose error prediction problem is defined as a transfer learning paradigm for the first time in this work, where theoretical knowledge of pose error distribution is embedded into the simulation domain through coarse calibration and a few samples are selected and measured by the MGS to form the high precision measurement domain. The knowledge of the simulation domain is transferred to the measurement domain in the form of model, data, and loss function by the proposed semi- supervised transfer learning (STL), which achieves a high accuracy of pose error prediction. The performance of ASTL was compared with other sample selection strategies and prediction algorithms on the four tasks constructed (POINTS, LINE, CURVE and SURFACE). The results show the accurate and efficient performance of ASTL, which is meaningful for future robotic high-precision applications.

Recommended citation: T. Zhang, F. Peng, X. Tang, R. Yan, C. Zhang, R. Deng, An active semi-supervised transfer learning method for robot pose error prediction and compensation, Eng. Appl. Artif. Intell., 128 (2024) 107476, https://doi.org/10.1016/j.engappai.2023.107476. https://doi.org/10.1016/j.engappai.2023.107476

An effective robotic processing errors prediction method considering temporal characteristics

Published in February 01, 2024

This paper presents a combined statistical principles and machine learning model that achieves high robot milling errors prediction accuracy, called PSO-ARIMA. It is an Auto-regressive Integrated Moving Average (ARIMA) model with milling force correction that has been optimized by the Particle Swarm Optimization (PSO). Compared to the other five existing algorithms, the proposed method has the highest prediction accuracy. The maximum MAE for pose errors prediction in the four validation tasks is only 0.021 mm and 0.011°, which meets the actual application requirements. It can efficiently and accurately accomplish online prediction of errors to improve the accuracy of robotic milling.

Recommended citation: R. DENG, X. TANG, T. ZHANG*, F. PENG, J. YUAN, R. YAN, An effective robotic processing errors prediction method considering temporal characteristics, Journal of Advanced Manufacturing Science and Technology, 0 (2024) 0, https://doi.org/10.51393/j.jamst.2024010. https://doi.org/10.51393/j.jamst.2024010

Quantification of uncertainty in robot pose errors and calibration of reliable compensation values

Published in March 13, 2024

In this work, the uncertainty in robot pose errors is focused on, to bridge the limitations of past uncertainty quantification methods such as assumption dependence and midpoint substitution, a joint prediction method for robot pose errors is proposed to achieve accurate estimation of point and intervals at the same time. In addition, to address the problem that most of the existing researches are based on the point prediction results for direct error compensation, which is heavily dependent on the point prediction accuracy, a reliable compensation value calibration strategy is proposed. The proposed strategy realizes further improvement of point prediction accuracy based on joint prediction. The proposed method is verified on three spatial trajectories of the ISO 9283-1998 standard, showing accurate joint prediction capability and reliable accuracy improvement. Through online compensation experiments, the pose errors are reduced by 90%, which promotes the application of robots in higher-precision scenarios.

Recommended citation: T. Zhang, F. Peng, R. Yan, X. Tang, R. Deng, J. Yuan, Quantification of uncertainty in robot pose errors and calibration of reliable compensation values, Robot. Comput.-Integr. Manuf., 89 (2024) 102765, https://doi.org/10.1016/j.rcim.2024.102765. https://doi.org/10.1016/j.rcim.2024.102765

An online prediction and compensation method for robot position errors embedded with error-motion correlation

Published in May 11, 2024

Error prediction and compensation are crucial requirements for improving robot accuracy. To this end, this paper introduces an advanced online method that integrates error-motion correlation for the precise prediction and compensation of robot position errors. The proposed method includes feature selection, model interpretability design, and compensation value smoothness. Analyzing the error characteristics and designing a model based on error-motion correlation reveals that the proposed method reduces position errors by 92% and 89% at a frequency of 250 Hz in two tasks with different working conditions. This reduction in position errors ensures smoother compensation, leading to a notable improvement in accuracy for high-precision robotics applications. Unlike existing research that focuses on mapping models, the proposed method prioritizes understanding error behavior, thereby resulting in a comprehensive approach for error management in robotic systems. This contribution is invaluable for advancing the field of precision robotics and ensuring reliable performances in various robotic tasks.

Recommended citation: T. Zhang, H. Sun, F. Peng, X. Tang, R. Yan, R. Deng, An online prediction and compensation method for robot position errors embedded with error-motion correlation, Measurement, (2024) 114866, https://doi.org/10.1016/j.measurement.2024.114866. https://doi.org/10.1016/j.measurement.2024.114866

A sparse knowledge embedded configuration optimization method for robotic machining system toward improving machining quality

Published in June 30, 2024

In recent years, robotic machining has become one of the most important paradigms for the machining of large and complex parts due to the advantages of large workspaces and flexible configurations. However, different configurations will correspond to very different system performances, influenced by the position-dependent properties. Therefore, the configuration optimization of robotic machining system is the key to ensure the quality of robotic operation. In response to the fact that little attention has been paid in current research to the effect of mapping model distribution differences on the optimization results, a sparse knowledge embedded configuration optimization method for robotic machining systems toward improving machining quality is proposed. The knowledge of theoretical model-based optimization in terms of stage, density and redundancy is embedded into high-fidelity data by three steps sparse and real measurement. Pre-training and domain adaptation fine-tuning strategies are used to reconstruct the real mapping model accurately. The reconstructed mapping model is re-optimized to obtain a more accurate system configuration. The effectiveness of the proposed method is verified by machining experiments on space segment parts. The proposed method reduces the absolute position error and machining error by 48.67 % and 28.73 %, respectively, compared to the current common theoretical model-based optimization. This is significant for more accurate and reliable robot system optimization. Furthermore, this work confirms the influence of mapping model distribution differences on the optimization effect, providing a new and effective perspective for subsequent research on the optimization of robotic machining system configurations.

Recommended citation: T. Zhang, F. Peng, X. Tang, R. Yan, R. Deng, S. Zhao, A sparse knowledge embedded configuration optimization method for robotic machining system toward improving machining quality, Robot. Comput.-Integr. Manuf., 90 (2024) 102818, https://doi.org/10.1016/j.rcim.2024.102818. https://doi.org/10.1016/j.rcim.2024.102818

An unsupervised prediction of robotic machining error for new tasks under historical tasks knowledge distillation

Published in January 01, 2025

Industrial robots have become an important machining equipment after machine tools in recent years, due to the advantages of, large workspace, and flexible work modes. However, the weak rigidity brought by the serial configuration leads the robot to shuffle from large errors. In addition, under the coupling of external time-varying dynamic loads and position-dependent characteristics, the robot machining error shows significant condition sensitivity. Data-driven becomes an effective mean. However, the performance of data-driven models is severely affected by the differences in working conditions，such as machining parameters, workpiece shape, etc. To this end, a new tasks unsupervised prediction method for robotic machining error prediction with historical knowledge distillation is proposed, in which the knowledge of historical tasks is distilled by a lightweight model, which achieves the transfer and reuse of domain knowledge, and improves the generalization ability of data-driven models for similar tasks. Based on eight historical tasks, robotic machining error prediction was performed for two new tasks, one with different parameters and the other with different shape. The effectiveness of the proposed method is fully verified by analyzing and comparing. Further, the scalable and maintainable digital twin system is initially constructed in order to better fulfill the needs of robotic manufacturing. The proposed method can provide a strong impetus to the development of knowledge-based intelligent manufacturing for robots.

Recommended citation: T. Zhang, F. Peng, X. Tang, Z. Yang, R. Yan, An unsupervised prediction of robotic machining error for new tasks under historical tasks knowledge distillation, Procedia CIRP, 133 (2025) 161-166, https://doi.org/10.1016/j.procir.2025.02.029. https://doi.org/10.1016/j.procir.2025.02.029

Spatial–temporal feature fusion for intelligent foreknowledge of robotic machining errors

Published in Jan 18, 2025

In this article, a spatial-temporal feature fusion model for intelligent foreknowledge of robot machining error is proposed. Different from the existing research on robot error or machining error, the proposed method pays attention to both the spatially-dependent ontological error characteristics of the robot and the time series characteristics of the robot machining process, and it is designed with the spatial-temporal feature extraction and the attention mechanism link for the above two characteristics, which realizes the feature extraction for different spatial arrangements and different workpiece shapes. Finally, after feature fusion and mapping, accurate prediction of robot machining errors is achieved.

Recommended citation: Zhang, T., Peng, F., Wang, J., Yang, Z., Tang, X., Yan, R., Zhao, S., & Deng, R. (2025). Spatial–temporal feature fusion for intelligent foreknowledge of robotic machining errors. Robotics and Computer-Integrated Manufacturing, 94, 102972, https://doi.org/10.1016/j.rcim.2025.102972. https://doi.org/10.1016/j.rcim.2025.102972

An uncertainty quantification and accuracy enhancement method for deep regression prediction scenarios

Published in Jan 20, 2025

In this study, a method for uncertainty quantification and accuracy enhancement, referred to as UQAE, was proposed for deep regression prediction scenarios, considering the relationship between point predictions and intervals. The method consists of two main components: a joint point-interval prediction module and a FIS-based accuracy enhancement module. These components comprehensively account for the point-interval similarity in joint predictions and the role of joint prediction results in calibrating point accuracy. To validate the advantages of the proposed method, three classes of datasets with nine tasks were constructed, and a fair comparison and analysis were conducted against current state-of-the-art methods. Based on the experimental results and analysis.

Recommended citation: Zhang, T., Peng, F., Yan, R., Tang, X., Yuan, J., & Deng, R. (2025). An uncertainty quantification and accuracy enhancement method for deep regression prediction scenarios. Mechanical Systems and Signal Processing, 227, 112394, https://doi.org/10.1016/j.ymssp.2025.112394. https://doi.org/10.1016/j.ymssp.2025.112394

UGP-KD An unsupervised generalized prediction framework for robot machining quality under historical task knowledge distillation for new tasks

Published in June 01, 2025

The influence mechanisms of robot errors by different working conditions and spatial ontology properties are explored in this paper. A spatial-temporal dual-view error prediction model is constructed for a single condition. Moreover, an innovative unsupervised generalized prediction strategy of machining error for new conditions under the historical task knowledge distillation of Multi-Teacher-Single-Student (MTSS) is proposed. This strategy enables the extraction and reuse of knowledge at three levels: teacher-teaching, student-learning, and generalized expansion. It also ensures the high-precision, lightweight, and high-efficiency prediction of machining error for unseen conditions.

Recommended citation: T. Zhang, F. Peng, Z. Yang, X. Tang, R. Yan, UGP-KD: An unsupervised generalized prediction framework for robot machining quality under historical task knowledge distillation for new tasks, Computers in Industry, 168 (2025) 104269, https://doi.org/10.1016/j.compind.2025.104269. https://doi.org/10.1016/j.compind.2025.104269

Digital twin-driven staged error prediction and compensation framework for the whole process of robotic machining

Published in September 12, 2025

Robotic machining has become another important machining paradigm after CNC machine tools. However, robot error has always been an important constraint in its progress towards high quality demand scenarios due to characteristics such as weak rigidity and pose dependence. Numerous scholars have carried out rich work around errors in robotic machining systems, and these studies have achieved excellent results in robot localization, trajectory continuous motion, and machining operations. However, due to the complexity of the robot machining system, the robot error has differentiated performance at different stages, and it is difficult to guarantee the global accuracy of the robot by focusing on and controlling a certain kind of error in a discrete manner. For this reason, a digital twin-driven staged error prediction and compensation framework for the whole robot machining process is constructed. In this framework, the whole process of robot machining is divided into three stages with significant differences: point planning, trajectory planning and material removal. And the error prediction function block in each stage is constructed for the error characteristics (distribution skew, error step, spatial-temporal coupling). For error compensation, a staged error compensation strategy is constructed from three aspects: offline point position, robot body and external three-axis platform, respectively. The constructed system was case-validated in the robotic machining of curved parts. All stages of the error prediction models show high prediction accuracy, and the excellent performance of the staged prediction models is verified by comparing with the classical prediction models. For the error compensation, the designed system is utilized to ensure that the robotic machining system provides a double guarantee on the robot end and the machining quality, the point position absolute error is controlled at 0.109 mm, the orientation error is controlled at 0.028°, the trajectory position error is controlled at 0.067 mm, the orientation error is controlled at 0.031°, and the final part machining error is controlled at 0.036 mm, which is almost approximates the repeatable positioning accuracy of the robot. The proposed framework realizes the system-level sensing and control of the robot machining system error, and provides a unified system framework for the subsequent research of related unit methods, which is conducive to promoting the development of robot machining to high-quality requirement scenarios.

Recommended citation: Zhang, T., Peng, F., Yang, Z., Tang, X., Yuan, J., & Yan, R. (2025). Digital twin-driven staged error prediction and compensation framework for the whole process of robotic machining. Journal of Manufacturing Systems, 83, 252-283, https://doi.org/10.1016/j.jmsy.2025.09.009. https://doi.org/10.1016/j.jmsy.2025.09.009

talks

teaching