Growing trends toward increased complexity and prolonged useful lives of engineering systems present challenges for system designers in accounting for the impacts of post-design activities on system performance (e.g., costs, reliability, customer satisfaction, environmental impacts). Examples of the post-design activities include manufacturing, condition monitoring, remaining life prediction, operations and maintenance, service logistics, as well as end-of-life options. It is difficult to develop accredited lifecycle system performance models because these activities only occur after the system is built and operated. Thus, system design and post-design decision-making have traditionally been addressed separately, leading to suboptimal performance over the system’s lifecycle.
With significant technological advances in computational modeling, simulation, sensing and condition monitoring, and machine learning and artificial intelligence, the capability of predictive modeling has grown exponentially over the past decade, leading to demonstrated benefits such as improved system availability and reduced operation and maintenance costs. Predictive modeling can bridge system design and post-design stages and provide an optimal pathway for system designers to effectively account for future system operations at the design stage. While predictive modeling of post-design activities potentially enables more holistic design decisions, there is a need to acquire improved knowledge on various aspects of this emerging topic, such as novel predictive modeling methodology for post-design activities, as well as novel design concepts incorporating predictive models into engineering design decision-making.
With 13 papers, this special issue brings together fundamental scientific contributions across different areas related to integrated design and operation of engineering systems (IDOES) with predictive modeling, such as novel modeling approaches, design concepts, as well as engineering design applications. Based on the research objective, the articles in the special issue are broadly grouped into five themes: (i) current literature on IDOES, (ii) novel predictive modeling approach for IDOES, (iii) new design concept for IDOES, (iv) new methodology for digital-twin design, and (v) novel engineering system applications of IDOES. In the following, the articles are briefly summarized within the five identified themes accordingly.
Current Literature on IDOES
In the article titled Towards Integrated Design and Operation of Complex Engineering Systems With Predictive Modeling: State-of-the-Art and Challenges, Liu et al. conducted a literature study on the integrated design and operation of engineering systems with predictive modeling, where predictive modeling approaches and strategies of integrating predictive models into the system design processes are categorized. Although predictive modeling has been handled from data-driven, statistical, analytical, and empirical aspects, and recent design problems have started to evaluate the lifecycle performance, there are still challenges in the field that require active investigation and exploration. Toward this end, this article provides a summary of future directions at the closure, encouraging research collaborations among various communities interested in the optimal system lifecycle design.
Novel Predictive Modeling Approach for IDOES
Multi-fidelity modeling and calibration are data fusion tasks that ubiquitously arise in engineering design. In the paper titled Data Fusion With Latent Map Gaussian Processes, Eweis-Labolle et al. addressed the critical need for general techniques that can jointly fuse multiple datasets with varying fidelity levels while also estimating calibration parameters. The authors introduced a novel approach that converts data fusion into a latent space learning problem using latent-map Gaussian processes, where the relations among different data sources can be automatically learned. By assimilating multiple datasets simultaneously using the proposed method, the prediction performance of multi-fidelity predictive models can be improved, as shown in the reported case studies.
Condition monitoring plays a crucial role in improving system failure resilience, preventing tragic consequences brought by unexpected system failure events and avoiding consequential increases in operation and maintenance costs. To integrate system designs with operations, a systematic decision-making framework is needed for system designers to assess the value of condition monitoring systems at the design stage, which would allow system design decisions on adopting monitoring systems to maximize the benefits. In the article titled Valuation of Continuous Monitoring Systems for Engineering System Design in Recurrent Maintenance Decision Scenarios, Liu et al. presented a decision-making framework to assess the value of condition monitoring systems based on the value of information. The proposed framework enables system designers to evaluate expected operation cost reductions under specific operation modes considering the effectiveness of continuous monitoring systems in predicting system failures.
In the paper titled Reinforcement Learning-Based Sequential Batch-Sampling for Bayesian Optimal Experimental Design, Ashenafi et al. presented a new sequential batch-sampling method for developing predictive models. The developed method, referred to as Bayesian sequential sampling via reinforcement learning for Bayesian optimal experimental design, can be used to optimize black-box and expensive-to-compute experiments or computer codes. The sampling algorithm developed in the article can select batches of queries while considering the entire budget for the design of experiments in hand, which retains the sequential nature while incorporating elements of reward based on tasks from the domain of deep reinforcement learning.
New Design Concept for IDOES
In the article titled Artificial Intelligence Design for Ship Structures: A Variant Multiple-Input Neural Network-Based Ship Resistance Prediction, Ao et al. developed a data-driven artificial intelligence (AI)-based predictive model to assist the ship hull design process. Specifically, an AI-based multiple-input neural network model was developed and implemented to realize a real-time prediction of the total resistance of the ship hull structure while avoiding inconsistent estimates from different types of design input parameters. It is demonstrated that the developed AI-based machine learning prediction tool can assist the ship hull design process by accurately estimating the total resistance of ship hulls in real time.
Path tracking error control is an essential functionality in developing autonomous vehicles to follow a planned trajectory, as significant path tracking errors could lead to a collision or affect the vehicle control algorithm significantly. While model-based control strategies are currently available, the bias of the baseline vehicle model may result in significant path tracking errors. In the article titled Bias-Learning-Based Model Predictive Controller Design for Reliable Path Tracking of Autonomous Vehicles Under Model and Environmental Uncertainty, Ren et al. presented a real-time bias-learning method coupled with model predictive control to improve the fidelity of a baseline vehicle model using a few experiments, so that the path tracking error can be reduced in real-time operation. In their presented study, Gaussian process regression and recurrent neural network were employed for bias learning, and their effectiveness was compared under different uncertainty scenarios.
In the article titled Design of Unmanned Cable Shovel Based on Multiobjective Co-Design Optimization of Structural and Control Parameters, Zhang et al. presented a multi-stage multi-objective co-design optimization strategy to jointly optimize the structure design and operations control of unmanned cable shovels considering excavation and loading processes, which have traditionally been performed at different stages, making it difficult to obtain the global optimal solution. Predictive models have been developed for point-to-point motion trajectory and the energy consumption in the working process, which are then integrated into the co-design framework to synchronously optimize the structural and control parameters, thereby improving mining efficiency and reducing energy consumption in unmanned excavation scenarios.
Novel Methodology for Digital-Twin Design
In the article titled Functional Modeling-Based Digital Twin Architecture Representation: An Instructional Example of a COVID-19 Breathalyzer Kiosk, Kotecha et al. presented a functional modeling-based representation of digital-twin architectures and illustrated it with an instructional example of a COVID-19 testing breathalyzer kiosk design. They provided a review of existing architectures and frameworks intended for use on product digital twin and proposed a modeling-based digital-twin architecture representation approach that potentially opens new venues of research helping to improve the design process for product digital twins.
In the article titled Probabilistic Digital Twin for Additive Manufacturing Process Design and Control, Nath et al. developed a new method for constructing a probabilistic digital twin for a laser power bed fusion (LPBF)-based additive manufacturing (AM) process that incorporates model uncertainty and process variability. The resulting digital twin using the developed method can thus be tailored for the individual part being produced using the AM process and used for probabilistic process parameter optimization and online, real-time adjustment of the LPBF process parameters, to control the porosity of the manufactured part.
In the article titled Design of Digital Twin Sensing Strategies via Predictive Modeling and Interpretable Machine Learning, Kapteyn et al. presented a new methodology for sensor placement and dynamic sensor scheduling decisions for digital twins. The digital-twin data assimilation is posed as a classification problem, and predictive models are used to train optimal classification trees that represent the map from observed data to estimated digital-twin states. In addition to providing a rapid digital-twin updating capability, the resulting classification trees yield an interpretable mathematical representation that can be queried to inform sensor placement and sensor scheduling decisions.
Novel Engineering System Applications of IDOES
In the article titled Mobility Prediction of Off-Road Ground Vehicles Using a Dynamic Ensemble of NARX Models, Liu et al. presented a surrogate modeling approach for predicting the mobility of off-road autonomous ground vehicles (AGVs), which is essential for AGV model-based mission planning especially in the early design stage. A dynamic ensemble of nonlinear autoregressive network with exogenous input (NARX) models over time was employed using synthetic vehicle mobility data of an AGV from a limited number of high-fidelity simulations. The presented case study demonstrated the advantages of the proposed predictive modeling method for AGV mobility prediction.
In the article titled Reliability-Based Multivehicle Path Planning Under Uncertainty Using a Bio-Inspired Approach, Liu et al. presented a novel bio-inspired approach for model-based multi-vehicle mission planning under uncertainty for off-road AGVs subjected to mobility reliability constraints in dynamic environments. Identifying a reliable path in uncertain environments is essential for designing reliable off-road AGVs considering post-design operations. An adaptive surrogate modeling method based on physics-based simulations was utilized to predict the vehicle state mobility reliability in operation, and subsequently, a bio-inspired approach called Physarum-based algorithm is used in conjunction with a navigation mesh to identify an optimal path satisfying a specific mobility reliability requirement.
In the article titled Dynamic Resource Allocation in System-of-Systems Using a Heuristic-Based Interpretable Deep Reinforcement Learning, Chen et al. developed a dynamic two-tier learning framework, based on deep reinforcement learning that enables dynamic resource allocation while acknowledging the autonomy of systems constituents for system of systems (SoS) applications. The two-tier learning framework that decouples the learning process of the SoS constituents from that of the resource manager ensures that the autonomy and learning of the SoS constituents are not compromised as a result of interventions executed by the resource manager. The authors applied the developed learning framework to a customized Open AI Gym environment and compared the results with baseline methods of resource allocation to show the superior performance across a different set of SoS key parameters.