The potential of artificial intelligence (AI) and machine learning (ML) in industrial settings is both exciting and daunting. With the ability to understand and predict equipment behaviour, these technologies promise to revolutionize maintenance and operational strategies. However, caution is necessary, as relying solely on these technologies can lead to significant risks.

Why Do Most Maintenance Transformations Fail?

As much as 70% of maintenance transformations fail. This high failure rate is often due to organizations being stuck in a cycle of small-scale pilots and lacking the readiness for complex AI-driven solutions.

Common pitfalls include:

1. Endless piloting without scaling.
2. Systems unprepared for predictive solutions.
3. Initiatives that, despite best efforts, fail to deliver sustainable outcomes.

Understanding AI’s Strengths and Limitations

AI excels in repetitive processes, such as fault detection in similar equipment across fleets. It also performs well in tasks like natural language processing and generating dashboard reports. A good example would be to start with numerous similar items, such as pipes or pumps, where AI can identify patterns and predict failures.

However, AI is not a cure-all. It's not effective in predicting rare or unique events, especially when there isn't sufficient data to establish a pattern. Key questions to consider before implementing AI include:

• Are the events frequent enough to predict?
• Are the failures similar and under consistent conditions?

The Myth of Unbiased Algorithms

AI requires a "ground truth"—clear, measurable outcomes to learn from. In maintenance, like in medicine, avoiding dysfunction means there's often no ground truth for failures, making it difficult for AI to learn effectively. A 2020 study by Griffiths et al. highlighted this issue, showing a lack of correlation between hydraulic faults and maintenance records.

Integrating Machine Learning into Your Business

Introducing AI is more than just a project—it's a major shift in business operations and culture. This shift requires careful planning, alignment with business strategies, and a robust governance model. Continuous monitoring and validation of AI models are crucial, even after deployment.

The Role of Human Decision-Making

Despite the power of AI, human oversight remains essential. Trust in the data and the analytics team is vital, as misplaced trust can lead to significant issues. Maintaining human involvement ensures that decisions consider broader business implications and ethical concerns.

Implications for Asset and Maintenance Managers

For asset and maintenance managers, the challenge is to integrate AI into the existing maintenance cycle: identify, plan, schedule, execute, complete, and analyse. AI can add value, particularly in the 'identify' stage of this cycle.

While AI and ML offer transformative potential, they also require a realistic understanding of their capabilities and limitations. Proper implementation involves not just technological changes but also significant cultural and organizational shifts. By carefully managing these factors, organizations can harness the power of AI to create more reliable and efficient operations.

This is an excerpt from a blog and discussion with Melinda Hodkiewicz, Professor of Engineering at the University of Western Australia.  For full and complementary access, please visit the MAINSTREAM Network.