Dr. Ming Zhang

Dr. Ming Zhang

Research Associate, College Of Engineering And Physical Sci, Aston University, Birmingham, UK

Publications

Dynamic Scheduling Method for Job-Shop Manufacturing Systems by Deep Reinforcement Learning with Proximal Policy Optimization

Published in Sustainability, 2022

With the rapid development of Industrial 4.0, the modern manufacturing system has been experiencing profoundly digital transformation. The development of new technologies helps to improve the efficiency of production and the quality of products. However, for the increasingly complex production systems, operational decision making encounters more challenges in terms of having sustainable manufacturing to satisfy customers and markets’ rapidly changing demands. Nowadays, rule-based heuristic approaches are widely used for scheduling management in production systems, which, however, significantly depends on the expert domain knowledge. In this way, the efficiency of decision making could not be guaranteed nor meet the dynamic scheduling requirement in the job-shop manufacturing environment. In this study, we propose using deep reinforcement learning (DRL) methods to tackle the dynamic scheduling problem in the job-shop manufacturing system with unexpected machine failure. The proximal policy optimization (PPO) algorithm was used in the DRL framework to accelerate the learning process and improve performance. The proposed method was testified within a real-world dynamic production environment, and it performs better compared with the state-of-the-art methods.

Recommended citation: Zhang M, Lu Y, Hu Y, Amaitik N, Xu Y. Dynamic Scheduling Method for Job-Shop Manufacturing Systems by Deep Reinforcement Learning with Proximal Policy Optimization. Sustainability. 2022; 14(9):5177. https://doi.org/10.3390/su14095177. https://doi.org/10.3390/su14095177

Cost Modelling to Support Optimum Selection of Life Extension Strategy for Industrial Equipment in Smart Manufacturing

Published in Circular Economy and Sustainability, 2022

Industrial equipment/machinery is an important element of manufacturing. They are used for producing objects that people need for everyday use. Therefore, there is a challenge to adopt effective maintenance strategies to keep them well-functioning and well-maintained in production lines. This will save energy and materials and contribute genuinely to the circular economy and creating value. Remanufacturing or refurbishment is one of the strategies to extend life of such industrial equipment. The paper presents an initial framework of cost estimation model based on combination of activity-based costing (ABC) and human expertise to assist the decision-making on best life extension strategy (e.g. remanufacturing, refurbishment, repair) for industrial equipment. Firstly, ABC cost model is developed to calculate cost of life extension strategy to be used as a benchmark strategy. Next, expert opinions are employed to modify data of benchmark strategy, which is then used to estimate costs of other life extension strategies. The developed cost model has been implemented in VBA-based Excel® platform. A case study with application examples has been used to demonstrate the results of the initial cost model developed and its applicability in estimating and analysing cost of applying life extension strategy for industrial equipment. Finally, conclusions on the developed cost model have been reported.

Recommended citation: Amaitik, N., Zhang, M., Wang, Z. et al. Cost Modelling to Support Optimum Selection of Life Extension Strategy for Industrial Equipment in Smart Manufacturing. Circ.Econ.Sust. (2022). https://doi.org/10.1007/s43615-022-00154-0. https://doi.org/10.1007/s43615-022-00154-0

Predictive Maintenance for Remanufacturing Based on Hybrid-Driven Remaining Useful Life Prediction

Published in Applied Sciences, 2022

Remanufacturing is an activity of the circular economy model whose purpose is to keep the high value of products and materials. As opposed to the currently employed linear economic model, remanufacturing targets the extension of products and reduces the unnecessary and wasteful use of resources. Remanufacturing, along with health status monitoring, constitutes a key element for lifetime extension and reuse of large industrial equipment. The major challenge is to determine if a machine is worth remanufacturing and when is the optimal time to perform remanufacturing. The present work proposes a new predictive maintenance framework for the remanufacturing process based on a combination of remaining useful life prediction and condition monitoring methods. A hybrid-driven approach was used to combine the advantages of the knowledge model and historical data. The proposed method has been verified on the realistic run-to-failure rolling bearing degradation dataset. The experimental results combined with visualization analysis have proven the effectiveness of the proposed method.

Recommended citation: Zhang M, Amaitik N, Wang Z, Xu Y, Maisuradze A, Peschl M, Tzovaras D. Predictive Maintenance for Remanufacturing Based on Hybrid-Driven Remaining Useful Life Prediction. Applied Sciences. 2022; 12(7):3218. https://doi.org/10.3390/app12073218. https://doi.org/10.3390/app12073218

Resonance Impedance Shaping Control of Hip Robotic Exoskeleton

Published in 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2021

The hip assistance robotic exoskeleton has been demonstrated as an effective device to assist elderly and disabled people with gait disorders. The assistance efficiency of these devices, however, is less optimized because the parameters in the active impedance control are manually designated. This paper presented a novel assistance control scheme to address the sub-optimal issue. This study poses that the assistance efficiency can be maximized by modifying the mechanical impedance to resonate with the muscle driving force, in which the human-exoskeleton coupling system is approximated with a second-order dynamical system. Based on this, the exoskeleton virtual stiffness is adaptively tuned to make the system intrinsic frequency align with the intended swing frequency. The proposed assistance control scheme demonstrated an increased assistance efficiency than the conventional active impedance control in a simulated study. Experiments that were managed on a newly custom-made hip assistance robotic exoskeleton also demonstrated strong evidence of improved gait kinematics with decreased muscle-skeleton efforts.

Recommended citation: T. Xue et al., "Resonance Impedance Shaping Control of Hip Robotic Exoskeleton," 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC), 2021, pp. 888-893, doi: 10.1109/SMC52423.2021.9659270. https://doi.org/10.1109/SMC52423.2021.9659270

Remaining Useful Life Estimation for Turbofan Engine with Transformer-based Deep Architecture

Published in 2021 26th International Conference on Automation and Computing (ICAC), 2021

With the development of information technology and sensors, the large industrial system has become a data-rich environment, which leads to the rapid development and application of deep learning for the remaining useful life prediction, especially for the turbofan engine. Currently, the deep architecture of CNN, LSTM have been used to address the RUL estimation of a turbofan engine. However, they are mainly focused on simulation degradation data. The new realistic run-to-failure turbofan engine degradation dataset has been published in 2021, which presents a significant difference from the simulation one. The main challenge is that the flight duration of each cycle is different, which will result in the current deep method hardly used for predicting the RUL for the practical degradation data. To tackle this challenge, we propose a novel Transformer-based model using guiding features to deal with the unfixed-length data. Besides, our G-Transformer model makes use of multi-head attention to access the global features from various representation subspaces. We conduct experiments on turbofan engine degradation data with variable-length input under practical flight conditions. Empirical results and feature visualization via t-SNE indicate the effectiveness of the G-Transformer model for RUL estimation of turbofan engines.

Recommended citation: Q. Ma, M. Zhang, Y. Xu, J. Song and T. Zhang, "Remaining Useful Life Estimation for Turbofan Engine with Transformer-based Deep Architecture," 2021 26th International Conference on Automation and Computing (ICAC), 2021, pp. 1-6, doi: 10.23919/ICAC50006.2021.9594150. https://doi.org/10.23919/ICAC50006.2021.9594150

Metric-based meta-learning model for few-shot fault diagnosis under multiple limited data conditions

Published in Mechanical Systems and Signal Processing, 2021

The real-world large industry has gradually become a data-rich environment with the development of information and sensor technology, making the technology of data-driven fault diagnosis acquire a thriving development and application. The success of these advanced methods depends on the assumption that enough labeled samples for each fault type are available. However, in some practical situations, it is extremely difficult to collect enough data, e.g., when the sudden catastrophic failure happens, only a few samples can be acquired before the system shuts down. This phenomenon leads to the few-shot fault diagnosis aiming at distinguishing the failure attribution accurately under very limited data conditions. In this paper, we propose a new approach, called Feature Space Metric-based Meta-learning Model (FSM3), to overcome the challenge of the few-shot fault diagnosis under multiple limited data conditions. Our method is a mixture of general supervised learning and episodic metric meta-learning, which will exploit both the attribute information from individual samples and the similarity information from sample groups. The experiment results demonstrate that our method outperforms a series of baseline methods on the 1-shot and 5-shot learning tasks of bearing and gearbox fault diagnosis across various limited data conditions. The time complexity and implementation difficulty have been analyzed to show that our method has relatively high feasibility. The feature embedding is visualized by t-SNE to investigate the effectiveness of our proposed model.

Recommended citation: Wang D, Zhang M, Xu Y, et al. Metric-based meta-learning model for few-shot fault diagnosis under multiple limited data conditions[J]. Mechanical Systems and Signal Processing, 2021, 155: 107510.. https://doi.org/10.1016/j.ymssp.2020.107510

A New Implementation of Digital Twins for Fault Diagnosis of Large Industrial Equipment

Published in Journal of Robotics and Mechanical Engineering Research, 2021

Refurbishment and remanufacturing play a vital role in the sustainability of the large industrial field, which aims at restoring the equipment that is close to the end of their life. The EU-funded project RECLAIM proposes new approaches and techniques to support these two activities in order to achieve saving valuable materials and resources by renewing and recycling the mechanical equipment rather than scraping them when they exceed the end of the lifetime. As the most critical part of predictive maintenance in RECLAIM, the fault diagnosis technique could provide the necessary information about the identification of the failure type, thus making suitable maintenance strategies. In this paper, we propose a novel implementation method that can combine the digital twins with the fault diagnosis of large industrial equipment. Experiment result and analysis demonstrate that the proposed framework performs well for the fault diagnosis of rolling bearing.

Recommended citation: M. Zhang, N. Amaitik, Y. Xu, et al. A New Implementation of Digital Twins for Fault Diagnosis of Large Industrial Equipment. J Robot Mech Eng. 2021;1: pp 1-7. https://doi.org/10.53996/2770-4122.jrme.1000103

Boosting Few-Shot Learning with Task-Adaptive Multi-level Mixed Supervision

Published in CAAI International Conference on Artificial Intelligence, 2021

In this paper, we propose a novel task-adaptive few-shot learning (FSL) method called Multi-Level Mixed Supervision (MLMS), which adapts a classifier specifically for each task by mixed supervision. Our method complements the supervised training with a multi-level unsupervised loss including the instance-level certainty term, set-level divergence term, and group-level consistency term. We further modify the set-level divergence term under the unbalanced prior situation where different classes of the unlabeled set contain different numbers of samples. Besides, we propose an approximate solution of minimizing our MLMS loss which is faster than the gradient-based method. Extensive experiments on multiple FSL datasets demonstrate that our method outperforms several recent models by an obvious margin on both transductive FSL and semi-supervised FSL tasks.

Recommended citation: Wang, D., Ma, Q., Zhang, M., Zhang, T. (2021). Boosting Few-Shot Learning with Task-Adaptive Multi-level Mixed Supervision. In: Fang, L., Chen, Y., Zhai, G., Wang, J., Wang, R., Dong, W. (eds) Artificial Intelligence. CICAI 2021. Lecture Notes in Computer Science(), vol 13070. Springer, Cham. https://doi.org/10.1007/978-3-030-93049-3_15 https://link.springer.com/chapter/10.1007/978-3-030-93049-3_15

Wasserstein distance guided adversarial imitation learning with reward shape exploration

Published in 2020 IEEE 9th Data Driven Control and Learning Systems Conference (DDCLS), 2020

The generative adversarial imitation learning (GAIL) has provided an adversarial learning framework for imitating expert policy from demonstrations in high-dimensional continuous tasks. However, almost all GAIL and its extensions only design a kind of reward function of logarithmic form in the adversarial training strategy with the Jensen-Shannon (JS) divergence for all complex environments. The fixed logarithmic type of reward function may be difficult to solve all complex tasks, and the vanishing gradients problem caused by the JS divergence will harm the adversarial learning process. In this paper, we propose a new algorithm named Wasserstein Distance guided Adversarial Imitation Learning (WDAIL) for promoting the performance of imitation learning (IL). There are three improvements in our method: (a) introducing the Wasserstein distance to obtain more appropriate measure in adversarial training process, (b) using proximal policy optimization (PPO) in the reinforcement learning stage which is much simpler to implement and makes the algorithm more efficient, and (c) exploring different reward function shapes to suit different tasks for improving the performance. The experiment results show that the learning procedure remains remarkably stable, and achieves significant performance in the complex continuous control tasks of MuJoCo.

Recommended citation: M. Zhang et al., "Wasserstein Distance guided Adversarial Imitation Learning with Reward Shape Exploration," 2020 IEEE 9th Data Driven Control and Learning Systems Conference (DDCLS), 2020, pp. 1165-1170, doi: 10.1109/DDCLS49620.2020.9275169. https://doi.org/10.1109/DDCLS49620.2020.9275169

Domain adaptation with multilayer adversarial learning for fault diagnosis of gearbox under multiple operating conditions

Published in 2019 Prognostics and System Health Management Conference (PHM-Qingdao), 2019

Deep learning has been widely developed to solve fault diagnosis issues, and it is becoming a crucial technology in the modern manufacturing industry. As an important transmission device of mechanical equipment, gearbox often runs at different speeds and loads, which may lead to changes in data distribution for the actual application. The cross-domain problem caused by the different data distribution may decline the performance of the fault diagnosis model based on deep learning. To overcome this challenge, a new domain adaptation method, named MAAN: Multilayer Adversarial Adaptation Networks, for fault diagnosis of gearbox running at multiple operating conditions. The basic framework of our MAAD is a deep convolutional neural network (CNN) and then an adversarial adaptation learning procedure is used for optimizing the basic CNN to adapt cross different domain. The results of the experiment demonstrate that MAAN has outstanding fault diagnosis and domain adaptation capacity, and it could obtain high accuracies for fault diagnosis of the gearbox with changing mode. For investigating the adaptability in this method, we use t-SNE to reduce the high dimension feature for better visualization.

Recommended citation: M. Zhang, W. Lu, J. Yang, D. Wang and L. Bin, "Domain Adaptation with Multilayer Adversarial Learning for Fault Diagnosis of Gearbox under Multiple Operating Conditions," 2019 Prognostics and System Health Management Conference (PHM-Qingdao), 2019, pp. 1-6, doi: 10.1109/PHM-Qingdao46334.2019.8943056. https://doi.org/10.1109/PHM-Qingdao46334.2019.8943056

A Deep Transfer Model With Wasserstein Distance Guided Multi-Adversarial Networks for Bearing Fault Diagnosis Under Different Working Conditions

Published in IEEE Access, 2019

n recent years, intelligent fault diagnosis technology with the deep learning algorithm has been widely used in the manufacturing industry for substituting time-consuming human analysis method to enhance the efficiency of fault diagnosis. The rolling bearing as the connection between the rotor and support is the crucial component in rotating equipment. However, the working condition of the rolling bearing is under changing with complex operation demand, which will significantly degrade the performance of the intelligent fault diagnosis method. In this paper, a new deep transfer model based on Wasserstein distance guided multi-adversarial networks (WDMAN) is proposed to address this problem. The WDMAN model exploits complex feature space structures to enable the transfer of different data distributions based on multiple domain critic networks. The essence of our method is learning the shared feature representation by minimizing the Wasserstein distance between the source domain and target domain distribution in an adversarial training way. The experiment results demonstrate that our model outperforms the state-of-the-art methods on rolling bearing fault diagnosis under different working conditions. The t-distributed stochastic neighbor embedding (t-SNE) technology is used to visualize the learned domain invariant feature and investigate the transferability behind the great performance of our proposed model.

Recommended citation: M. Zhang, D. Wang, W. Lu, J. Yang, Z. Li and B. Liang, "A Deep Transfer Model With Wasserstein Distance Guided Multi-Adversarial Networks for Bearing Fault Diagnosis Under Different Working Conditions," in IEEE Access, vol. 7, pp. 65303-65318, 2019, doi: 10.1109/ACCESS.2019.2916935. https://doi.org/10.1109/ACCESS.2019.2916935

Research on variational mode decomposition in rolling bearings fault diagnosis of the multistage centrifugal pump

Published in Mechanical Systems and Signal Processing, 2017

Rolling bearing faults are among the primary causes of breakdown in multistage centrifugal pump. A novel method of rolling bearings fault diagnosis based on variational mode decomposition is presented in this contribution. The rolling bearing fault signal calculating model of different location defect is established by failure mechanism analysis, and the simulation vibration signal of the proposed fault model is investigated by FFT and envelope analysis. A comparison has gone to evaluate the performance of bearing defect characteristic extraction for rolling bearings simulation signal by using VMD and EMD. The result of comparison verifies the VMD can accurately extract the principal mode of bearing fault signal, and it better than EMD in bearing defect characteristic extraction. The VMD is then applied to detect different location fault features for rolling bearings fault diagnosis via modeling simulation vibration signal and practical vibration signal. The analysis result of simulation and experiment proves that the proposed method can successfully diagnosis rolling bearings fault.

Recommended citation: Zhang M, Jiang Z, Feng K. Research on variational mode decomposition in rolling bearings fault diagnosis of the multistage centrifugal pump[J]. Mechanical Systems and Signal Processing, 2017, 93: 460-493. https://doi.org/10.1016/j.ymssp.2017.02.013

Dynamic analysis of integrally geared compressors with varying workloads

Published in Shock and Vibration, 2016

Integrally geared compressors are characterized by compact and high efficiency machines, which are widely used in modern processing industries. As an important part of integrally geared compressors, a geared rotor-bearing system exhibits complicated dynamic behaviors. When running at rated speeds, a coupling system likely produces resonance with an adjusted workload, and a critical load phenomenon occurs. The dynamic coefficients of bearings, axial force and torque, and gear meshing stiffness vary with workload because of the interaction between rotors. In this study, a dynamic model of a geared rotor-bearing system influenced by the dynamic coefficients of bearings, axial force and torque, and gear meshing stiffness is developed. The dynamic responses of the coupling system are calculated and analyzed by using a typical five-shaft integrally geared compressor as an example. The effects of different parameters on the dynamic behaviors of the proposed system are also considered in the discussion. The geared rotor-bearing system is further investigated to examine the failure mechanism of the critical load.

Recommended citation: Ming Zhang, Zhinong Jiang, Jinji Gao, "Dynamic Analysis of Integrally Geared Compressors with Varying Workloads", Shock and Vibration, vol. 2016, Article ID 2594635, 13 pages, 2016. https://doi.org/10.1155/2016/2594635 https://doi.org/10.1155/2016/2594635