Predictive Maintenance Scheduling Research Articles

Multi-objective predictive maintenance scheduling models integrating remaining useful life prediction and maintenance decisions

This paper presents two novel data-driven multi-objective predictive maintenance scheduling models that integrate remaining useful life (RUL) prediction into maintenance planning. First, we propose a deep learning ensemble model that includes a convolutional neural network and bidirectional long short-term memory network with a temporal self-attentional mechanism and Bayesian optimization method to predict system RUL effectively. Then, two novel multi-objective mixed-integer linear programming (MILP) models are developed based on the system-predicted RUL to deal with the system predictive maintenance problem, which aims to minimize the total maintenance completion time and maintenance-related costs simultaneously. To solve this problem, we design an iteration ϵ-constraint method to achieve Pareto solutions. Meanwhile, a fuzzy logic method is proposed to recommend a preferred Pareto solution for decision-makers. Finally, the aircraft engine C-MAPSS dataset from NASA validates that the effectiveness of the proposed data-driven multi-objective predictive maintenance scheduling models are effective. For the predictive maintenance of 20 aircraft engines, both the maintenance completion time and maintenance cost are reduced by more than 40%.

Predictive Maintenance Scheduling for Aircraft Engines Based on Remaining Useful Life Prediction

Predictive Maintenance Scheduling for Aircraft Engines Based on Remaining Useful Life Prediction

Health Monitoring System for CNC Machine

This paper introduces a novel Health Monitoring System (HMS) tailored for CNC machines, addressing the critical need for maintaining their optimal performance and preventing unexpected breakdowns in modern manufacturing settings. The system integrates various sensors and data acquisition methods to continuously monitor key parameters like temperature, vibration, and tool wear. By employing advanced data analytics and machine learning algorithms, the HMS can swiftly identify anomalies in real-time, facilitating proactive maintenance and minimizing operational downtime. Additionally, the system features a user- friendly interface for visualizing machine health status and creating predictive maintenance schedules. Experimental validation conducted on a CNC machining center validates the efficacy and reliability of the developed HMS in enhancing machine efficiency, prolonging equipment lifespan, and curbing maintenance expenses. In summary, this Health Monitoring System offers a robust solution for ensuring the seamless operation and longevity of CNC machines in modern manufacturing environments.

An experimental hybrid customized AI and generative AI chatbot human machine interface to improve a factory troubleshooting downtime in the context of Industry 5.0

Throughout industrial revolutions, equipment downtime mitigations have been one of the ultimate goals of most factories. Several tools, such as human machine interface (HMI) alarming systems or predictive maintenance schedules, assist in reducing system downtime but still depend on the operators’ ability to swiftly retrieve, understand, and efficiently act upon reported failures. We propose the design of a hybrid experimental artificial intelligence (AI) and generative AI chatbot HMI that effectively extracts factory equipment conditions that are useful for troubleshooting and predictive maintenance analysis. We achieve these functions by feeding experimental factory-monitored data to the customized chatbot application tool running in its back-end, a Langchain agent linked to the OpenAI GPT-\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$-$$\\end{document}3.5 language model (LM) via OpenAI APIs. We design our chatbot front-end with Streamlit, an open-source web app. In the context of I5.0, our chatbot HMI uses personalized natural language, English, to interact with the operator, making the information extraction more understandable. We also integrate the generative AI capability of the GPT 3.5 LM that augments the factory data based on the loaded format to create a larger dataset for additional tasks like machine learning modelling. The experimental results show the accuracy of our customized chatbot HMI when retrieving data based on specific prompts and the advantages of a reduced troubleshooting time compared to operations in traditional factories, which are highly dependent on supervisors’ interventions. Our study provides a valuable example of upgrading standard factory HMIs to I5.0-capable ones by implementing customized AI and generative AI chatbots within operational industrial environments.

XR and digital twins, and their role in human factor studies

Digital twins (DT) are synchronized clones that mirror their physical counterparts at all times, enabling real-time monitoring, analysis, decision-making, and planning for optimized operations. Digital twins can transcend traditional two-dimensional interfaces by incorporating VR/AR/MR (XR) technology, providing immersive, intuitive, and interactive representations of systems, assets, and processes. Furthermore, real-time data from sensors, simulations, and other sources can be integrated into the XR-enabled digital twins, leading to better and more intuitive understanding and to more efficiently monitor, analyze, and maintain complex systems such as nuclear assets. Immersive, interactive, multi-player XR capabilities embedded in DTs will allow spatially accurate and more realistic representation, leading to improved risk assessment, optimized and predictive maintenance scheduling, enhanced situational awareness, and more effective communication among interdisciplinary teams. By combining the strengths of XR and digital twins, nuclear facilities can achieve heightened safety, operational efficiency, and decision-making accuracy. Marrying XR technology with digital twins is also likely to extend the utilization of digital twins to optimize those aspects of the design and operation of nuclear assets that involve human beings–specifically in human factors engineering. Training can also be significantly enhanced if DTs are linked with XR technology. These systems may also be used to assess human performance through human factors engineering for the safety analysis of nuclear assets. A specific example is assessing human performance in semi-autonomous nuclear assets or operating multiple nuclear assets. After briefly reviewing digital twins of nuclear systems from the perspective of XR technology, this paper summarises our work in the nuclear energy space on VR/AR/MR and how these can be integrated into the newly evolving DTs of nuclear assets. The paper also describes the potential use of such systems in optimizing the design and operations of nuclear systems. As XR technology advances, its symbiotic relationship with digital twins can significantly reshape the landscape of nuclear operations and asset management.

Research on integrated scheduling of equipment predictive maintenance and production decision based on physical modeling approach

Equipment performance deteriorates continuously during the production process, which makes it difficult to achieve the expected effect of production decisions made in advance. Predictive maintenance and production decisions integrated scheduling aim to rationalise maintenance activities. It has been extensively researched. However, past studies have assumed that faults obey a specific probability distribution based on historical data. It is difficult to analyse equipment that is brand new into service or has poor historical failure data. Thus, in this paper, we construct a twin model of a device based on a physical modelling approach and tune it to ensure high fidelity of the model. Degradation curves were created based on equipment characteristics and developed maintenance activities.Develop an integrated scheduling model for predictive maintenance and production decisions with the goal of minimising maximum processing time. An improved genetic algorithm is used to solve the problem optimally. Finally, apply a practical scenario to verify the effectiveness of the proposed method.

Battery Energy Storage Capacity Estimation for Microgrids Using Digital Twin Concept

Globally, renewable energy-based power generation is experiencing exponential growth due to concerns over the environmental impacts of traditional power generation methods. Microgrids (MGs) are commonly employed to integrate renewable sources due to their distributed nature, with batteries often used to compensate for power fluctuations caused by the intermittency of renewable energy sources. However, sudden fluctuations in the power supply can negatively impact battery performance, making it challenging to select an appropriate battery energy storage system (BESS) at the design stage of an MG. The cycle count of a battery in relation to battery stress is a useful measure for determining the general health of a battery and can aid in BESS selection. An accurate digital replica of an MG is required to determine the required cycle count and stress levels of a BESS. The Digital Twin (DT) concept can be used to replicate the dynamics of the MG in a virtual environment, allowing for the estimation of required cycle numbers and applied stress levels to a BESS. This paper presents a Microgrid Digital Twin (MGDT) model that can estimate the required cycle count and stress levels of a BESS without considering any unique battery type. Based on the results, designers can select an appropriate BESS for the MG, and the MGDT can also be used to roughly estimate the health of the currently operating BESS, allowing for cost-effective predictive maintenance scheduling for MGs.

A Tractable Failure Probability Prediction Model for Predictive Maintenance Scheduling of Large-Scale Modular-Multilevel-Converters

Modular-multilevel-converters (MMCs) are vital components in direct current transmission networks. Predictive maintenance scheduling of MMCs requires estimations of the failure probabilities of MMCs during a period of time in the future. Particularly, the predicted future failure probabilities are influenced by two main factors, the mission profiles of the MMCs and the maintenance decisions on the MMCs during the prediction period. This paper proposes a failure probability prediction model (FPM) for MMCs by considering these two factors. First, the expectations of the failure probabilities of the components for all the scenarios of mission profiles are obtained. Second, in predictive maintenance scheduling problems, the decisions to perform the maintenance actions are represented by binary variables. When the number of submodules is very large, using the binomial probability form currently used in reliability engineering to express the “r-out-of-n” failure probability of arms of the MMCs is intractable. Thus, this paper proposes a tractable form (T-form) in FPM by observing that the submodules on one arm are homogeneous. Furthermore, an approximation method, i.e., clustering and assignment (C&A), is proposed to reduce the computation times for calculating the parameters needed by the proposed T-form. Then, we perform a case study that assesses the accuracy and computation time of the C&A approach. The results show that the accuracy of the C&A approach is high and that the computation time is reduced significantly compared with the accurate method. We also show that the computation time for solving the predictive maintenance scheduling problem can be reduced hugely by using the T-form instead of the binomial probability form.

OPTIMIZING WIND FARM PERFORMANCE USING MACHINE LEARNING

The rapid expansion of wind energy as a sustainable power source has necessitated the adoption of advanced technologies to optimize wind farm performance. This study explores the application of machine learning (ML) algorithms to enhance the efficiency and maintenance scheduling of wind turbines. By leveraging predictive analytics, wind farm operators can significantly improve operational performance, reduce downtime, and extend the lifespan of turbines. Machine learning models can analyze vast amounts of data generated by wind turbines, including sensor readings, weather conditions, and historical performance metrics. These models identify patterns and predict potential failures before they occur, allowing for proactive maintenance and reducing the risk of unexpected breakdowns. Predictive maintenance scheduling, informed by ML algorithms, ensures that turbines are serviced at optimal times, minimizing disruption and maximizing energy production. The study examines various ML techniques such as regression analysis, neural networks, and decision trees, which are applied to predict turbine performance and detect anomalies. Case studies demonstrate the effectiveness of these techniques in real-world scenarios, highlighting improvements in energy output and maintenance efficiency. For instance, regression models can predict the power output based on wind speed and direction, while neural networks can detect subtle changes in vibration data that may indicate mechanical issues. Moreover, the integration of machine learning into wind farm operations supports the development of a more resilient and adaptive energy infrastructure. Real-time data analysis enables dynamic adjustments to turbine settings, optimizing performance under varying environmental conditions. This adaptability not only enhances the efficiency of individual turbines but also contributes to the overall stability and reliability of the wind farm. The findings of this study underscore the transformative potential of machine learning in renewable energy management. By implementing predictive algorithms, wind farm operators can achieve significant gains in efficiency, reliability, and cost-effectiveness. This research advocates for the broader adoption of machine learning technologies in the wind energy sector, paving the way for more sustainable and economically viable renewable energy solutions. Optimizing wind farm performance through machine learning represents a promising avenue for advancing the efficiency and sustainability of wind energy. Predictive algorithms offer a powerful tool for enhancing maintenance scheduling and operational performance, driving the future of renewable energy innovation. Keywords: ML, Wind Farm Performances, Predictive Algorithms, Efficiency, Maintenance Scheduling.

Predictive Maintenance and Upgrade Scheduling for Production Line under Multi-period Product-service Paradigm

In the rapidly developing product–service paradigm, a multi-unit manufacturing system will be leased many times during life cycle. Besides maintenance programming for a single leasing period, an upgrade policy is essential to achieve life-cycle profit optimization during sequential periods. This article presents a framework integrating opportunistic maintenance scheduling and upgrade decision for the leased system throughout its life cycle based on a real-time prognostic method. With machine-level prognosis updating, system-level opportunistic predictive maintenance is dynamically achieved by leasing profit-saving calculations during each leasing period. Between successive leasing contracts, the optimal upgrade levels are dynamically determined through upgrade profit-saving maximization. A numerical example of a leased manufacturing line demonstrates the long-term economic effectiveness of this policy. The proposed framework extends traditional opportunistic maintenance policies in both the machine-level prognosis and the system-level upgrade decision. This framework can be applied to a broad range of leased systems with longer service lifetimes.

Speed-Dependent Eigenmodes for Efficient Simulation of Transverse Rotor Vibration

Accurate, computationally efficient simulations enable engineers to design high-performing, cost-efficient, lightweight machines that can leverage models of predictive controls and digital twin predictive maintenance schedules. This study demonstrates a new speed-dependent eigenmode method for accurately and efficiently simulating shaft transverse vibrations. The method involves first independently computing shaft eigenmodes over a range of operating speeds, then correlating the eigenmodes across the different speeds during compilation, and finally adjusting modal properties gradually in accordance with a lookup method during simulation. The new method offers several distinct advantages over the traditional static eigenmodes and Craig-Bampton methods. The new method maintains accuracy over a large range of shaft rotation speeds whereas the static eigenmodes method does not. The new method typically requires fewer modal degrees of freedom than the Craig-Bampton method. Whereas the Craig-Bampton method is limited to modeling changes at the boundaries, the new method is suitable for modeling changing body properties as well as boundary-based changes. For this paper, a fluid-bearing-supported 10 MW direct-drive wind turbine drive shaft is tested virtually in a simulation model developed in Simscape™ Driveline™. Using the simulation statistics, this study compares the accuracy and computational efficiency of the speed-dependent eigenmode method to the traditional finite lumped element, static eigenmode, and Craig–Bampton methods. This paper shows that the new method simulates the chosen system 5 times faster than the traditional lumped mass method and 2.4 times faster than the Craig-Bampton method.

Collaborative production and predictive maintenance scheduling for flexible flow shop with stochastic interruptions and monitoring data

Flexible flow shop problem (FFSP) has been extensively researched in recent years. However, machine deterioration and stochastic production interruptions noticeably affect the regular job process in practice, incurring great cost. It is significant to incorporate the condition-based predictive maintenance scheduling and the production modification for interruptions with traditional FFSP. Nevertheless, this issue is rarely investigated in the existing FFSP studies. To solve the FFSP with maintenance and stochastic interruptions (FFSP-MSI), this paper proposes a collaborative optimization policy of production and maintenance (COPPM) by considering stochastic interruptions and monitoring data utilization for total cost reduction. For maintenance scheduling, a real-time prognosis updating method is leveraged based on monitoring data to capture the individual machine degradations. Assisted by the prognosis output, the optimal maintenance intervals are dynamically derived through a cost rate model. For production optimization, a variable neighborhood search (VNS) algorithm is presented to obtain the nearly- optimal initial production plan and the production modification for stochastic interruptions caused by machine failures and random jobs. This proposed COPPM comprehensively determines the cost-effective maintenance cycle, job order and machine selection for each job. Computational experiments of the wind turbine blade process are presented to certify the economic effectiveness of COPPM. Compared with the traditional production and maintenance scheduling method, the COPPM achieves 13.68 % total cost reduction in average under various problem scales.

Early Fault Warning and Identification in Condition Monitoring of Bearing via Wavelet Packet Decomposition Coupled With Graph

It has been well known that the fault of rolling element bearings (REBs) is one of the biggest causes for machine breakdown; thus, the early detection and type identification of a fault during REBs successive operations is necessary to help users take predictive maintenance action and/or schedule in order to avoid major machine failures. Wavelet packet decomposition (WPD) is a widely used technique to analyze vibration signal for health monitoring of REBs, but the resulting wavelet packet coefficients (WPCs) are still in high dimensionality, making them impossible for the direct use in practical usage. Keeping along the research line of high-level analysis for WPD enhancement, this article presents a new dynamic modeling approach, called graph-modeled wavelet packet coefficients (GMWPCs), that integrates WPD and graph theory in order to extract the correlation information between WPCs. By means of an adaptive input weight fusion, the GMWPCs can automatically confirm the weight of the frequency sub-band where the fault-induced information is more evident. By virtue of GMWPCs, a new two-phase framework for early warning detection and fault identification is proposed finally, which is able to not only detect the time location of the fault in its very early stage but also identify the fault type. We conducted different experiments to validate the early warning detection and fault identification separately. Experimental results, outperforming the state of the arts, verify the adequacy and effectiveness of the proposed framework and meanwhile reveal its appropriate use for real engineering applications.

Integrated Multi-objective Optimization of Predictive Maintenance and Production Scheduling: Perspective from Lead Time Constraints

For the integrated optimization of job-shop production scheduling and predictive maintenance, this paper fully considers such constraints as product delivery time and changing machine failure rate, and establishes a multi-objective optimization model aiming to minimize the processing cost and the product processing time. The model includes the changing machine failure rate into the integrated optimization of job-shop production scheduling and predictive maintenance, and enables the prediction of the machine state according to the processing time of the current job, laying the basis for the decision-making of the machine activity and the reasonable and effective production planning. In addition, the non-dominated sorting genetic algorithm (NSGA)-II was designed to solve the proposed model. The algorithm performance was improved through the operator crossover and mutation by the simulated binary crossover algorithm (SBX). The proposed strategy was verified through a case study.

Augmented model-based framework for battery remaining useful life prediction

Traditional, model-based approaches for predicting the remaining useful life (RUL) of a rechargeable battery cell simply update and extrapolate a mathematical model which describes the evolution of the cell’s capacity fade trend. These approaches are straightforward but tend to break down when the capacity fade trend changes over the cell’s lifetime. To retain the desirable properties of model-based prediction approaches (uncertainty quantification, long-term accuracy, limited physical meaning) and improve their overall accuracy in RUL prediction, we augment empirical model-based prediction with data-driven error correction. Our approach decomposes the task of RUL prediction into two steps: 1) Offline training of data-driven models for RUL error correction and 2) Online data-driven correction of model-based RUL prediction. The approach is evaluated on five datasets consisting of 237 cells: 1) three open-source datasets, 2) one proprietary dataset, and 3) a simulated out-of-distribution dataset. Results show that data-driven error correction effectively reduces root-mean-square-error by 40% and mean uncertainty calibration error by 34% compared to a model-based approach alone. The proposed approach is also shown to be more conservative in its uncertainty estimates than a purely data-driven RUL prediction approach. Special attention is given to ensure the initial model-based uncertainty estimates are propagated through the data-driven error correction model and considered in the final RUL prediction. The enhanced uncertainty quantification of our approach makes it suitable for deployment in an online predictive maintenance scheduling framework.

Optimization for a Joint Predictive Maintenance and Job Scheduling Problem With Endogenous Yield Rates

While job scheduling problems have been studied extensively, scheduling problems with endogenous yield rates that may be affected by predictive maintenance is not thoroughly investigated. In this study, we consider the optimization of a joint predictive maintenance and job scheduling problem for the minimization of total shortage penalty. As maintenance may be conducted to raise machine yield rates, machine production rates and job processing times become endogenous, and the optimization problem is different from traditional scheduling problems. We formulate a mixed integer program for this problem and develop a heuristic algorithm based on Tabu search. We demonstrate the effectiveness of our algorithm through numerical experiments and a way of estimating the yield declining rate with industry defect and maintenance records. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This work is motivated by the real need of our industry collaborator, red electronics manufacturer. Every morning, the manufacturer chooses up to three out of eleven photolithography machines to conduct maintenance. Conducting maintenance helps raise machine yield rates to decrease the number of defects in expectation. However, the production schedule of some jobs must be postponed, and delay and shortage may arise. The decision is thus to schedule jobs as well as maintenance to find a balance between yield loss and shortage loss. We help the manufacturer by formulating an optimization model and develop an algorithm to solve the model. The algorithm may be applied to similar cases when one needs to schedule maintenance and production processes at the same time.

Stochastic Resource Optimization over Heterogeneous Graph Neural Networks for Failure-Predictive Maintenance Scheduling

Resource optimization for predictive maintenance is a challenging computational problem that requires inferring and reasoning over stochastic failure models and dynamically allocating repair resources. Predictive maintenance scheduling is typically performed with a combination of ad hoc, hand-crafted heuristics with manual scheduling corrections by human domain experts, which is a labor-intensive process that is hard to scale. In this paper, we develop an innovative heterogeneous graph neural network to automatically learn an end-to-end resource scheduling policy. Our approach is fully graph-based with the addition of state summary and decision value nodes that provides a computationally lightweight and nonparametric means to perform dynamic scheduling. We augment our policy optimization procedure to enable robust learning in highly stochastic environments for which typical actor-critic reinforcement learning methods are ill-suited. In consultation with aerospace industry partners, we develop a virtual predictive-maintenance environment for a heterogeneous fleet of aircraft, called AirME. Our approach sets a new state-of-the-art by outperforming conventional, hand-crafted heuristics and baseline learning methods across problem sizes and various objective functions.

Alarm-based predictive maintenance scheduling for aircraft engines with imperfect Remaining Useful Life prognostics

The increasing availability of condition monitoring data for aircraft components has incentivized the development of Remaining Useful Life (RUL) prognostics in the past years. However, only few studies consider the integration of such prognostics into maintenance planning. In this paper we propose a dynamic, predictive maintenance scheduling framework for a fleet of aircraft taking into account imperfect RUL prognostics. These prognostics are periodically updated. Based on the evolution of the prognostics over time, alarms are triggered. The scheduling of maintenance tasks is initiated only after these alarms are triggered. Alarms ensure that maintenance tasks are not rescheduled multiple times. A maintenance task is scheduled using a safety factor, to account for potential errors in the RUL prognostics and thus avoid component failures. We illustrate our approach for a fleet of 20 aircraft, each equipped with 2 turbofan engines. A Convolution Neural Network is proposed to obtain RUL prognostics. An integer linear program is used to schedule aircraft for maintenance. With our alarm-based maintenance framework, the costs with engine failures account for only 7.4% of the total maintenance costs. In general, we provide a roadmap to integrate imperfect RUL prognostics into the maintenance planning of a fleet of vehicles.

Machine learning for predictive maintenance scheduling of distribution transformers

PurposeThe purpose of this paper is to describe a methodology that has been set up to schedule predictive maintenance of distribution transformers at Cauca Department (Colombia) using machine learning.Design/methodology/approachThe proposed methodology relies on classification predictive model that finds the minimal number of distribution transformers prone to failure. To verify this, the model was implemented and tested with real data in Cauca Department Colombia.FindingsThe implementation of the methodology allows a saving of 13% in corrective maintenance expenses for the year 2020.Originality/valueThe proposed model is an effective decision-making tool that provides an ideal solution for preventive maintenance scheduling problems for distribution transformers.

A predictive maintenance model for optimizing production schedule using deep neural networks

Industry 4.0 (I4.0) provides connectivity, data volume, new devices, miniaturization, inventory reduction, personalization, and controlled production. In this new era, customization and data availability are essential to generate information that allows decision-making. The possibility of predicting the need for maintenance in the future and using this information for other processes is one of the challenges of the manufacturing process. In this context, this article presents a model to optimize maintenance and production schedules predictively and automatically, called Predictive Maintenance (PdM) & Schedule (PdMS). Usually, these solutions are addressed separately in literature. We aim to improve the machines’ Remaining Useful Life (RUL) prognosis based on sensor telemetry and operating information. We also aim to predict if the machines will be part of the production schedule, considering the available resources. We propose a model that describes data engineering, validation, and normalization. It also describes our approach to create and combine degradation indices using similarity patterns that allow noisy data to identify time-based failures. This approach allows for the use of this type of prediction in scheduling problems. We compare several models based on deep neural networks (DNN) and recurrent neural networks (RNN), using criteria based on visual analysis, errors, regression coefficient R2, and accuracy. The best result, RMSE=8.789,MSE=77.253,MAE=2.262,R2=0.848,Accuracy=92.22, allowed for predicting failures with five-day anticipation, which avoids downtime and makes the intended integration possible.