With exponential digitalisation, asset management systems are becoming increasingly sophisticated, bringing new data sources and a data-driven approach. Weather data, asset information and sensor data from equipment can be integrated and given to machine learning algorithms to predict and prevent disruptions and failures. Ever sophisticated approaches enabled by new available data can help businesses to reduce downtimes, avoid reactive maintenance costs and lower preventive maintenance costs and raise customer satisfaction levels.
Traditionally, the approach to maintaining assets is focused on preventative maintenance with strict maintenance regimes, standard inspection cycles and renewal policy based on the lifetime of the asset calculated using theoretical engineering expertise. While such approaches are in general effective at reducing downtime, they can also lack efficiency and flexibility, and are often costly.
A machine learning approach to PAM allows businesses to design maintenance policies based on real time data and to improve decision making around asset renewal and maintenance. Engineering knowledge still plays a central role, captured in the model definition stage and through feature engineering, for example identifying key features and understanding different types of failures.
Further use of sensors in equipment and the availability of increasing amounts of data is forming ideal environments for machine learning to thrive and drive business value from PAM systems. However, organisations will only consider investing in machine learning if they can see immediate Return on Investment (ROI): it is difficult to prove value from machine learning models without creating a strong case that includes savings calculations based on evidence.
The question arises: how do we prove the value of machine learning with PAM systems? And what are the business benefits companies can hope to see?
Businesses need to wake up to the potential of machine learning driven PAM systems in terms of utilising asset data. The challenge is that the value of machine learning models is not always clear, and the cost benefit analysis can be challenging to conduct. Consequently, stakeholders are often reluctant to invest in machine learning driven PAM systems.
It’s here that data scientists and PAM experts can play a key role, once they acknowledge the importance of proving real business value to the stakeholders. It is not enough to provide a good machine learning model that accurately predicts the probability of a failure. Data scientists need to develop actionable insight for the business.
Experts and data scientists must communicate the benefits of machine learning through developing optimised maintenance plans, given the importance of ROI evaluation. These optimised maintenance plans are the maintenance policy for renewal and repair of assets based on probability of a failure and prioritization, and they are calculated using machine learning techniques. They also need to consider business specific requirements including maintenance costs and the number of available repair crews. This actionable insight can bring measurable benefit to companies, and can be effective in reducing communication gaps between data scientists and stakeholders.
Achieving success with PAM
Machine learning based PAM systems can drive significant transformational and business value in an array of industries. Here’s how:
A train manufacturer
A worldwide train manufacturer wanted to improve the servicing of trains and minimise downtime to raise customer satisfaction. The company wanted to use PAM models to understand the root cause of failures, and to give its Operations teams time to react to train disruptions, reducing downtime. With efficient machine learning models, maintenance costs would decrease and the manufacturer’s supply chain would be optimised through its ability to order parts when needed.
The PAM models involved were built using a variety of data sources including sensor data from different components, historical maintenance data, and weather data. The models were used to help understand the most critical components and influence factors of downtime. This insight helped to inform failure risk management, and over time the team could design predictive models to respond to this risk in an automated way.
A logistics company
A logistics company wanted to be able to predict container ship engine component defects using sensor data. Predicting and preventing ship engine component failures saves the shipping companies unproductive time worth millions of dollars. Traditionally, real time streaming data requires an expert who understands the engine enough to be able to monitor it. Alerts can be raised based on rules to detect “abnormal” states of an engine based on engineering knowledge, thus there is some level of automation to traditional preventive maintenance of this asset.
Machine learning models were developed to predict failures with a ten-day lead time. The models were trained on historical sensor data including around vibrations and the temperature of the engine. The developed models enable the automated processing of huge volumes of sensor data, as well as accurate reporting of the probability of an engine component failure within ten days.
The outputs of the PAM model can be used to setup an automated system for raising alerts every time the engine is at risk of failure. The key benefit of implementing such a system is the reduction of unexpected downtime by raising preventive alerts and minimising manual inspection costs by relying more on sensor data – and applying this data-driven approach.
Through PAM with machine learning models, organisations can adopt a data-driven approach to managing asset failure risk while improving efficiency and testing current engineering assumptions surrounding assets. To reveal its full potential however, PAM must be adopted more widely. Thus, it is essential to display its benefits to business leaders clearly by designing optimised maintenance plans that can demonstrate interpretable and actionable insight.
Marat is a data scientist at Think Big Analytics who focuses on delivering value for businesses using machine learning and data science. He has been involved in analytics projects across a number of industries such as Capital Markets, Insurance, Transport and Utilities. His areas of interest include predictive modelling, anomaly detection and more recently he has been developing his expertise in deep learning. In addition to his technical skills, Marat focuses on the positive real-world outcomes of data science by contributing to project definition, solution design and project management.