Digital transformation is crucial for businesses seeking to stay ahead in an ever-evolving market. While most C-suite executives are keen to embrace new technologies, efforts at end-to-end digital transformation sometimes result in technology spending that cannot deliver the expected value. These dynamics have recently played out in most predictive maintenance efforts by energy companies. Today, energy companies find themselves well-positioned to begin accelerating the impact of predictive maintenance transformation by resolving the issues that have held their solutions back.

Asset-intensive industries have been working on predictive maintenance solutions for nearly a decade. Of all the technologies that define Industry 4.0 (IoT, digital twins, big data, cloud computing, blockchain, AR/VR, etc.), many industry leaders feel that AI/ML-based predictive maintenance solutions will most significantly impact their bottom lines (by reducing variable cost) and top lines (by enhancing asset availability). Seeking easy-to-deploy predictive maintenance capabilities, many energy companies have turned to standard COTS solutions. 

Most of these predictive maintenance efforts are failing to achieve the intended results. These efforts are hamstrung by foundational challenges like insufficient data availability, poor data quality, integration issues, variable operating conditions, inappropriate predictive algorithms, ineffective data collection methods, lingering cybersecurity concerns, and change management challenges. Any one of those issues can impact the outcomes of the underlying AI/ML models, and many companies need help on multiple fronts. 

For companies dissatisfied with the state of their plug-and-play predictive maintenance outcomes, it’s time to take a step back and re-validate their approach. To transform their data into reliable and effective predictive maintenance interventions, companies must focus on four critical predictive maintenance enablers: comprehensive data collection, automated model training, advanced analytics, and cross-functional collaboration. 

Comprehensive Data Collection 

Many companies have been unsatisfied with their predictive maintenance solutions because their older equipment does not provide the data required to enable predictive maintenance. 

When companies turn to the current standard COTS solutions on the market, they generally find that they can effectively support most of their assets. But often, the available data tags from older pieces of equipment don’t match and satisfy the templates available within the COTS products. In an industry where asset lifespans are measured on a scale of decades rather than months, there is a massive amount of out-of-scope legacy equipment which were built before predictive maintenance was even a possibility. In a 10-year-old plant, it’s reasonable to expect that 40% of the assets will not have sensors capable of supporting standard COTS solutions. 

Some organizations might accept this lack of instrumentation (and therefore visibility) on a significant subset of assets. But a better approach would identify alternative sets of data tags (both upstream and downstream in the plant process circuit) that show a strong correlation with the missing parameters. This customized extrapolation approach to data ingestion can often gain a surprising amount of clarity on the state of legacy assets, improving failure prediction and anomaly detection. 

Yes, this approach requires a bespoke plant-by-plant implementation rather than a plug-and-play COTS product. However, in the long run, the direct costs and downstream impacts of sudden stoppages, equipment breakdowns, and suboptimal performance will be much more painful than enabling a customized and comprehensive approach to data collection.  

Efficient Automated Model Training

Theoretically, it is easy to create predictive AI models, and all standard solutions come with pre-built predictive maintenance ML algorithms. However, the same algorithm may perform differently across asset groups. Geographical differences, the nature of the relevant sensors, allowable operating windows, and even variations in the operating philosophies of individual plants and other assets can impact the output of AI models. 

To achieve reliable predictive maintenance across all assets and plants, enterprises need an approach to AI model training that is automated and efficient, taking into account all relevant variables. The most important prerequisites are selecting the optimal training period lengths and establishing a co-relation matrix among all sensors. An automated range-selection framework should be built to avoid manual training errors, but the selection has to be manually verified before feeding the model. Meanwhile, a co-relation matrix is critical for providing the operating contexts leading to high precision and intelligent results. 

Advanced Analytics

Installing new sensors is often the most direct way to capture data relevant to predictive maintenance. However, upgrading assets with new sensors can involve untenable production stoppages. And even in cases where plants have recently made the hard decision to install new sensors or smarter equipment, they may only have a few months of data — not nearly enough to train and run a robust predictive model.

In these cases, and also for new capital projects with no operational history, companies can increasingly rely on advanced analytics capabilities like GenAI and process simulation models to generate synthetic data, simulate every failure scenario, and resolve data availability challenges. 

Cross-functional Collaboration

It’s common for elegant data-driven solutions to stumble simply because the enterprise isn’t fully prepared to leverage the new solution. In the case of predictive maintenance, change management needs to ensure that employees across the company can collaborate to act quickly on predictive insights. Change management can build alignment between global technology and business teams, and extend that alignment to plant-level implementation teams. 

Formal organizational change management initiatives will help ensure proper training, mindset change, and process adoption. Regarding technology, predictive solutions must be thoughtfully integrated with ERP/APM applications to create responsive feedback mechanisms. Predictive maintenance alerts should create notifications in the CMMS system to trigger timely in-person site inspections and broader maintenance strategy changes. Without a tight data-to-action feedback loop (and a collaborative culture committed to advancing that tight feedback loop), companies will fail to realize the full potential of their predictive solutions.

When Predictive Maintenance Works

A recent Wipro engagement with a petrochemical client helped transform their data analytics landscape and clearly illustrates the concrete benefits of predictive maintenance.

The client faced several troubling and interrelated challenges: rising maintenance expenses, unexpected breakdowns, an erosion of asset availability, and a lack of enterprise-level visibility on what was going wrong. The client sought to enable AI-based predictive analytics and real-time prescriptive recommendations.  

The client worked with Wipro to deploy AI/ML-based failure prediction across 45 plants and over 12,000 assets. The anomaly and failure prediction models achieved more than 85% accuracy — a massive win in the context of predictive maintenance. With these new models, the client has realized an additional 10,000 hours of equipment availability and expects to reduce spare parts and labor by $5 million every year moving forward. Meanwhile, business users benefit from a single easy-to-use dashboard that acts as a real-time decision support system for alert monitoring, asset health status checks, operational efficiency, and plant-to-plant comparisons, which has led to improved reporting and visibility.

Past predictive maintenance interventions have underwhelmed many asset-intensive companies. Fortunately, predictive maintenance is now reaching a stage of robust maturity. The AI models have improved, data best practices are well-understood, and the implementation pathways are clearer. Energy companies can and should confidently invest in prescriptive maintenance, knowing that a customized implementation approach can rapidly deliver concrete business value. 

About the Authors

Tanmoy Naskar
Managing Consultant, Downstream Energy Consulting

Drawing on his significant experience with asset performance management (APM) and enterprise asset management (EAM), Tanmoy works with downstream energy clients to drive technology-driven improvements in the maintenance and reliability space. His expertise spans RCM, FMEA, RCA, reliability analytics, lifecycle modeling, Intelligent Asset Management, spare parts optimization, connected workers and operator rounds, predictive and prescriptive maintenance, remote operations/maintenance, and real-time asset lifecycle tracking. He has been in the energy consulting space for more than fifteen years, providing thought leadership to global companies seeking to modernize their core foundation and drive the digital transformation of assets across the value chain.

Arnab Sarkar
Global Practice Head, Downstream Energy Consulting

Arnab is an adept thought leader with more than 29 years of experience across multiple industries. He has spent more than a decade in the energy industry, involved in the commissioning of greenfield refinery and petrochemical complexes, stabilizing plant operations and maintenance, and driving new capital projects. For the last 17 years, he has been helping energy customers build capabilities and unique solutions, driving business and technology transformation to continuously improve plant operations efficiency, process automation, asset performance management, process safety, and decarbonization.