Industrial automation research increasingly brings together control theory, optimization, artificial intelligence and physical-system modeling. The Mitsubishi Electric Research Laboratories (MERL) 2018–2019 annual report offers a view of how these disciplines can connect fundamental research with development needs across a broad industrial portfolio, including factory automation equipment, power systems, transportation and building technologies.
For engineers and technical decision-makers, the report’s main takeaway is not a single product or implementation recipe. It is the range of methods that can support complex systems: model predictive control, nonlinear and fault-tolerant control, data-driven analysis, robotics, and simulation of interacting physical processes. These approaches can help teams reason about system behavior, constraints and uncertainty before translating research into engineered applications.
Industrial automation research spans multiple disciplines
MERL describes its work as application-motivated basic research and advanced development in areas including physical modeling and simulation, signal processing, control, optimization and artificial intelligence. Its researchers collaborate across technical groups and with Mitsubishi Electric business units. That structure reflects a practical reality in automation: system challenges rarely fit neatly within one discipline.
A manufacturing system, for example, can involve equipment dynamics, process constraints, sensing, communications and supervisory decisions. Improvements in one layer may depend on a better model of another. Control algorithms need information about system behavior, while optimization methods need a clear representation of operational goals and constraints. Research that connects these pieces can inform the development of more capable industrial systems.
Model predictive control and optimization for constrained systems
Model predictive control (MPC) is a recurring area in the report. In general terms, MPC uses a model to evaluate possible future system behavior and choose control actions while accounting for constraints. MERL staff research included constrained and nonlinear control, networked control systems, stochastic systems and optimization algorithms, with applications noted in areas such as factory automation, automotive and aerospace systems, logistics and process control.
These topics matter when a system must coordinate several objectives or operate within physical and operational limits. Research into efficient optimization can also be relevant where control problems are large-scale or computationally demanding. The report identifies work on algorithms for large-scale and mixed-integer nonlinear programs, including applications in power grids, transportation and model-based process control.
For automation teams, the practical consideration is to match the method to the problem. A control approach must account for the system’s dynamics and constraints, and its computational requirements must suit the intended application. The report describes research directions, rather than specifying a particular implementation or performance outcome.
Physical-system modeling supports simulation and fault detection
Accurate models are foundational to analysis, simulation and control. MERL’s work in multi-physical systems included modeling and simulation of thermo-fluid systems and electromechanical energy conversion, as well as research on numerical methods and fault detection. These systems combine physical effects that may interact over time, making their behavior difficult to understand through isolated component models alone.
Researchers also worked on transient modeling frameworks for thermo-fluid systems and on reduced-order models. A reduced-order model aims to represent important system behavior with less computational complexity than a more detailed model. In suitable applications, such models can support faster analysis, although the report does not claim a particular speed improvement for a specific industrial deployment.
For plant and engineering teams, this area of research highlights the value of connecting simulation to the physical system and its operating conditions. Models can help investigate dynamic behavior and inform control design; fault-detection research, in turn, addresses the challenge of identifying when system behavior warrants attention.
Robotics, machine learning and data-driven methods
The report also points to work at the intersection of robotics, estimation and machine learning. Research interests included motion planning, workspace analysis for collaborative robots, robotic vision, and data-driven methods for control and system identification. System identification uses observed system behavior to develop or refine a model, while estimation infers relevant system states from available measurements.
These techniques are relevant to automation tasks where a robot must interpret its environment or where a system’s behavior is not fully captured by a fixed model. MERL’s staff research also covered anomaly detection, time-series analysis and machine learning. Such methods can help analyze complex data, but their usefulness depends on how well the problem, available measurements and operating context are defined.
Artificial intelligence also appeared in the report’s account of an award-winning computer-vision-based river water-level detection system developed with Mitsubishi Electric colleagues in Japan. The system was fielded as part of the MAISART AI technology platform. While this is not a factory automation example, it illustrates how computer vision and deep learning research can be applied to monitoring tasks.
Turning research into engineering priorities
The report describes MERL’s mission as generating technology and intellectual property in areas important to Mitsubishi Electric, and working with the company’s organizations to support new and improved products. Its business portfolio includes factory automation equipment such as programmable logic controllers, inverters, servo motors and processing machines, alongside systems in energy, buildings, transportation and communications.
For system integrators and plant leaders, the broad research portfolio suggests several useful questions when evaluating an automation challenge:
- What needs to be modeled? Identify the physical dynamics and interactions that influence performance or control.
- Which constraints matter? Clarify operational limits and objectives before selecting an optimization or control strategy.
- What can the available data support? Assess whether sensing and historical data are suitable for estimation, anomaly detection or learning-based methods.
- How will the method fit the system? Consider computational needs, existing control architecture and the path from research concept to engineering implementation.
Conclusion: connect theory to operational needs
Industrial automation research can contribute across the system lifecycle, from physical modeling and simulation to constrained control, optimization, robotics and data analysis. MERL’s 2018–2019 report demonstrates the breadth of disciplines being explored within one corporate research organization and their potential relevance to industrial sectors. The practical implication for automation decision-makers is to frame projects around the system’s actual dynamics, constraints and data—and then choose research-informed methods that fit those needs.