Predictive Maintenance: Expectations Vs Reality

Did you know that predictive maintenance is not a novel concept? Its roots can be traced to the times of the Second World War, when scientist CH Waddington made an interesting observation of “positive harm” being caused to the Royal Air Force bombers, due to the planned and preventive maintenance processes. It was studied that the planned maintenance rituals were indeed responsible for creating unplanned failures. This phenomenon was termed “The Waddington effect”. It resulted in the change of maintenance processes to align with the equipment conditions and their usage patterns – heralding the initiation of the predictive maintenance that we know today.

Fast forward – Industry 4.0

Industry 4.0 and the Fourth Industrial Revolution are often referred to interchangeably. The massive impact of this revolution is primarily due to the amalgamation of physical and digital entities. According to Klaus Schwab, who concocted the term “The Fourth Industrial Revolution” describes it as “evolving at an exponential rather than a linear pace” and terms it as a distinct one in terms of “velocity, scope, and systems impact”.

The unprecedented surge in the data along with an improved ecosystem of sensors, storage, network technologies, and remarkable progress made in Artificial Intelligence has resulted in the wider adoption of predictive maintenance across the industrial domains such as manufacturing, oil & gas, mining, Aerospace, etc.

Challenges for the Vendors

Many PDM (Predictive Maintenance) providers have been trying to penetrate an already crowded marketplace, without having an understanding of the unique problems being faced in the particular domain. Various issues have resulted like integration issues with the legacy systems, treating the process industry challenges as typical data science and Big Data challenges, similar to other domains such as finance, healthcare, and retail.

As per a McKinsey study, less than 30% of pilot engagements are ready for scaling and a staggering 84% of companies are languishing in pilot projects. This “pilot purgatory” primarily is due to a wide gap of understanding between industrial companies and technology vendors.

Big Data Approach Vs Little Data Reality

Predictive maintenance use cases involve more than mere data science know-how. Contrary to what many believe, it’s not a Big Data problem where millions of labeled data points can be made available to train the models with prescribed methods. In a real industrial scenario, massive failures are usually quite atypical, making the failure data points insufficient for making predictions. Gartner has termed such data as “Little Data”. Not to forget the other variables like environment, dynamic asset behaviors which may follow usual patterns over time. Furthermore, every machine is different, despite being from a similar brand and performing the same functions.

The asset monitoring data coming from sensors and PLCs is voluminous and can be technically termed as Big Data. However, because of the low failure incidents, finding a definitive and straightforward pattern is daunting. In most industrial plants, device variations exist – where a group of devices produces many data points, and another group of devices produce just a limited data from the PLCs. The primary challenge here is to align the varied frequencies of data and still build consistent and prescriptive analytics that can cover all the equipment of a plant.

The Big Data approach works best in context-rich systems, whereas industrial asset failures are rather covert in nature without a clearly established pattern. This makes conventional machine learning algorithms insufficient in predicting asset failures.

User is the Hero at Eugenie

A PDM system should essentially be tailor-made to address the needs of maintenance and reliability staff who prefer insights in an actionable manner. Therefore, the user experience can make or break a PDM project. If the advanced analytics cannot deliver the insights which are understandable by the operational personnel, all the efforts, and resources invested in technology may transpire fruitlessly. We have seen a myriad of in-house projects where massive investments were made in developing data science models and dashboards, only to find out a shocking lack of user engagement.

At Eugenie, we have conducted extensive user research, involving thousands of maintenance and reliability professionals to exclusively understand their problems, mindsets, goals, and lifestyles as well.

We know that the super-busy maintenance professionals have usually a very short time during the start of their shift to spot the machines, among the thousands of others, that need their maximum attention. Also, most operational staff do not wish to spend a lot of time studying asset performance graphs and deciphering insights. Yet, many technology vendors provide generic dashboards, without taking into account the specificity of the operators’ jobs.

Additionally, maintenance and reliability professionals have an abundance of acquired knowledge through experience and efficacy which must reflect in a predictive maintenance solution.

Explainable AI – An Operator’s Ally

We’ve always followed the approach of human-centered AI, where complex insights integrate seamlessly with the diagnostic procedures through an intuitive user interface and simple language. Our decision intelligence framework combines technology, user experience as well as OEM’s standard operating procedures to generate accurate predictions and insights. With Eugenie, operators can engage effectively by storing their past experiences with the assets, thus enriching Eugenie’s robust machine database.

Our Explainable AI framework is created as an ally and a guide for operational professionals. Despite dealing with asset data and cutting-edge technology, the first priority at Eugenie is ALWAYS our users. This, along with ultra-flexible and scalable architecture has made us a preferred choice of many top-tier industrial organizations.

Eugenie’s patented AI-based products comprising of Auto-ML platform, combined with its rich database of machines and a human-centered framework makes it a truly distinguished product. Unlike other solutions, Eugenie can generate actionable insights even in the absence of asset failure data points through powerful point-anomaly detection and deep learning methods. With Eugenie, asset health can be monitored without the need to manually analyze each sensor-level data. Instead, Eugenie’s auto-ML platform can diagnose machine health with asset-level monitoring without user intervention. Thus, the entire plant can be covered with machines ranging from high to low criticality. The results are delivered in the form of insights that lead to speedy action.

To know more about Eugenie’s offerings, get in touch with us.

Additionally, we are also committed to helping industrial companies to reduce their carbon footprints. Read our Sustainability Series to find out how we are leveraging AI to create a sustainable future.