With continued global market growth and an increasingly competitive environment, manufacturing industry is facing challenges and desires to seek continuous improvement. This effect is forcing manufacturers to squeeze every asset for maximum value and thereby calls for high-equipment effectiveness, and at the same time flexible and resilient manufacturing systems. Maintenance operations are essential to modern manufacturing systems in terms of minimizing unplanned down time, assuring product quality, reducing customer dissatisfaction, and maintaining advantages and competitiveness edge in the market. It has a long history that manufacturers struggle to find balanced maintenance strategies without significantly compromising system reliability or productivity. Intelligent maintenance systems (IMS) are designed to provide decision support tools to optimize maintenance operations. Intelligent prognostic and health management tools are imperative to identify effective, reliable, and cost-saving maintenance strategies to ensure consistent production with minimized unplanned downtime. This article aims to present a comprehensive review of the recent efforts and advances in prominent methods for maintenance in manufacturing industries over the last decades, identifying the existing research challenges, and outlining directions for future research.

Gearboxes are critical transmission components in the drivetrain of wind turbine, which have a dominant failure rate and the highest downtime loss in all wind turbine subsystems. However, load variations of wind turbine gearbox are far from smooth and usually nondeterministic, which result in inconsistent data distributions. To solve the problem, a novel performance degradation assessment and prognosis method based on maximum mean discrepancy is proposed to test the difference between data distributions and extract the characteristics of multi-source working conditions data. Besides, the increase in sensors will bring more difficulties to establish prediction models in real-world scenarios due to different installation locations. In view of this, a transfer learning strategy called joint distribution adaptation is utilized to adapt data distribution between multi-sensor signals. Nevertheless, the presence of background noise of wind turbine signals restricts the applicability of these algorithms in practice. To further reduce the distribution difference, a novel criterion is proposed to evaluate and measure the data distribution difference between known and tested working conditions based on the witness function of maximum mean discrepancy. The application and superiority of proposed methodology are validated using a wind turbine gearbox life-cycle test data set. Meanwhile, model comparison and cross-verification are conducted between conventional and proposed prediction models. The results indicate that the proposed method has a better performance in performance degradation assessment for wind turbine gearbox.

Modeling of dynamic manufacturing processes with slow drift using data-driven approaches is challenging because most data-driven models are trained by off-line data. In this paper, we propose to track slow drift of manufacturing processes using an online Bayesian Auto-Regression eXogenous (ARX) model with time-variant parameters. The model parameters are trained off-line and are updated online using Bayes’ rule. To avoid the frequent online model update, a Sample Importance (SI) test is proposed to screen online samples and only the sample that passes SI test is incremented to the VM model. The SI test decides the importance of the sample by collectively considering sample freshness, model prediction error and prediction uncertainty. Furthermore, the SI test is applied to off-line Data Base (DB) and an iteration algorithm is devised for off-line sample selection. The off-line sample selection algorithm can effectively reduce the sample redundancy in off-line DB while maintaining the model performance. To validate the effectiveness of the proposed method, Prognostics and Health Management (PHM) data challenge 2016 dataset is employed to predict material removal rate of chemical-mechanical planarization process. The validation results indicate that the proposed method outperforms existing state-of-art approaches in literature such as Just-in-Time (JIT) and deep learning models.

As vast amounts of data are saved, hard drive failure prediction is critical to reducing the cost of data loss and backup. Most existing studies used to detect the anomalous status of a hard drive using self-monitoring, analysis, and reporting technology (SMART) attributes, and then predicted whether the drive would have an impending failure. However, as most researchers focus on a specific model of hard drives, existing studies do not apply to different models due to the model-to-model variations. Furthermore, as anomaly detection algorithms were the focus in the literature, feature selection methods were less compared. This article proposes an evaluation methodology to compare feature selection methods and anomaly detection algorithms for hard drive failure prediction. It can quickly select the optimal algorithms for a specific model of drives. It contains an evaluation mechanism to assess feature selection methods from the perspectives of performance and robustness and assess the performance, the robustness, the efficiency, and the generalization of anomaly detection algorithms. Experiments on two data sets are implemented for validation, and the proposed method can achieve better performance than the existing approaches in the literature.