Fault diagnosis and planning optimisation within an e-maintenance framework

Systems and equipment must operate at increasingly higher levels of efficiency to compete at any production or operational level, nationally or internationally, not being possible a couple of decades ago. Improvements in quality, shorter response times and continuous changes in the demand for products and services generate higher performance needs in operation and maintenance.

Although in recent years this evolution has led to great advances, and research has been done to increase the availability and reliability of machinery, there is a long way to reach a complete optimization in many fields of application.

The main objective of this research work is to demonstrate the potential for improvement that techniques and methodologies related to prescriptive analytics can provide in industrial maintenance applications. The technologies developed can be grouped into three areas:

E-maintenance and interoperability

This section is particularly relevant to support strategies related to continuous predictive maintenance (on-line) using the latest technologies on the market. The continuous cost reduction provides clear opportunities for improvement in maintenance processes. Predictive technologies have evolved in recent years and important advances are being made in mechanical, thermal, electromechanical applications and more currently in electrical and electronic systems.

E-maintenance is fundamentally related to the development of collaborative and intelligent platforms that allow the integration of new smart sensors, communications systems, standards and protocols, concepts, storage and analysis methods, etc. that improve our range of possibilities and offer us the possibility of following a trend in the optimization of assets and processes, and interoperability among systems.

Fault diagnosis

The Bayesian Networks together with other information capture methodologies used in engineering provides automation of diagnosis and prediction of failures. Machine learning techniques are widely used for the same purpose, but they have the disadvantage that they need a wide range of data to enhance a reliable model, and still there is a risk of excessive coupling to the system with which the tests are performed.

In any case, machine learning techniques are more focused on the detection of specific failures, while the purpose of this section, like the rest of the thesis project, is not focused on solving a specific problem, but rather on the methodology that must be taken into account and the steps to follow to design a diagnostic system in the absence or scarcity of data, which happens many times.

Simulation of strategies and planning optimisation

The main objective of this section is to focus on technologies that allow optimizing maintenance strategies, either with more reliable designs or by improving maintenance decisions. Cost-effectiveness analysis is key because indicates if any benefit or competitive advantage can be achieved using an adequate maintenance strategy, especially with a predictive maintenance target. The use of cost-effectiveness simulations in this area helps decision-making when selecting a suitable maintenance strategy for the asset.

In addition, by using optimization algorithms we can improve maintenance planning, reducing the time and costs to perform the tasks in an asset fleet. During the thesis project, a multi-objective algorithm was developed based on the use of an Estimation of Distribution Algorithm (EDA) to optimize maintenance plans. The concept of prescriptive analytics is fully evident in this section based on the combination of predictive actions (e.g. Failure prediction), simulation of the cost-benefit of different related decisions (e.g. different maintenance options) and the search for the best decision (or the set of decisions for an asset fleet).