Artificial intelligence (AI) tools are being implemented everywhere, including in the industrial maintenance sector.
Industrial entities have long tried to shift away from reactive or preventive maintenance (PM) processes. With the right technologies, they can adopt predictive maintenance (PdM) strategies that maximize asset lifecycles and reduce the risk of unplanned downtime. But how can AI support and boost this approach?
Overall, the future of AI in industrial maintenance is bright. However, another major shift is required to unlock the full potential of artificial intelligence technologies in maintenance. Here’s a look at how far these tools have come, where they’re headed and how your organization can use artificial intelligence to gain an edge.
The Rise of AI in Industrial Maintenance
Artificial intelligence has existed in some form or another for decades. However, its widespread use has recently kicked into overdrive. The question is: why now?
The rapid adoption of AI in industrial maintenance and other fields can be chalked up to three key developments:
- Computing Power: The advancement of processing units has allowed AI models to digest vast amounts of data more efficiently
- Data Availability: The internet and IoT devices have created an explosion of data
- Research and Development: Continuous investment in AI has led to a cascade of breakthroughs
Artificial intelligence gains momentum with each remarkable new advancement. Today, AI is a viable tool for improving industrial maintenance practices. However, industrial entities have barely scratched the surface of what predictive AI tools are capable of.
How AI Turned Mainstream
For decades, only PhDs and researchers had access to artificial intelligence models. These early models weren’t especially useful for real-world applications. I-care and other solutions providers helped change that by developing custom, niche use cases that addressed specific business challenges. This approach led to deployment of AI-driven solutions, which became accessible to limited industrial end-users considered to be early-adopters in industry. These solutions mostly were worked out with specialists within the organization. While they were definitely not widespread or democratically accessible, they were valuable nonetheless.
Generative AI (GenAI) like ChatGPT have changed everything. Suddenly, business leaders could see the potential of AI firsthand without prior training and expertise in the topic. They had a chance to experiment with artificial intelligence tools and use them to make routine daily tasks easier.
How AI Fits Into the Process-Optimization Equation
Artificial intelligence has already proven its worth by enhancing several key industrial processes. It excels in the following areas:
- Predictive Analytics: Reviewing real-time sensor data to detect patterns indicating impending failures
- Optimizing Process Efficiency: Adjusting operating parameters to reduce energy consumption and increase output
- Optimizing Maintenance Scheduling: Ensuring that maintenance is performed with minimal production impact
While traditional AI has successfully optimized these and other workflows, GenAI introduces a new dimension. GenAI can assist in writing maintenance procedures, analyzing historical logs and suggesting solutions to complex mechanical problems. GenAI can also be used to write specific program code to solve delimited use-cases.
The possibilities are worth getting excited about. However, these new use cases also present some unique challenges.
The Challenges of AI in Maintenance
Using generative artificial intelligence for predictive maintenance is an exciting proposition. That said, industrial entities will need to overcome some major hurdles first.
In a recent interview, Tom Rombouts, Director of Reliability and Data-Driven Solutions at I-care, summarized these challenges as follows:
Reliability
AI analyzes large volumes of data to generate results. If data on which it’s trained is biased or otherwise unreliable, the quality of the results will suffer. Therefore, all AI outputs must be carefully reviewed to ensure accuracy and trustworthiness.
Quality Control
Ensuring that AI-generated content, such as maintenance instructions, is sound and safe is vital. Incorrect procedures established due to GenAI hallucinations could lead to serious safety hazards and additional unplanned downtime. Subject matter experts must review all content to verify its accuracy and applicability without falling into the trap of relying on “confidently incorrect” GenAI output.
Change resistance
The I-care team firmly believes that digital transformation can’t be just digital. Implementing new technologies without providing staff with the training and confidence to use those tools will only produce underwhelming results.
But don’t mistakenly undervalue AI. It truly is “the change” of our generation!
Employees who are caught off guard by the change AI initiatives bring may develop strong feelings of resistance. The outcome will likely be a diminished ROI and prolonged time to value, both of which are counterproductive to an industrial organization’s maintenance goals.
What Needs to Happen Next
After learning about the efficiency benefits of artificial intelligence, you may be tempted to turn AI loose and let it solve all of your maintenance headaches. However, that’s a recipe for disaster. Without the right quality-control measures in place, generative AI can do just as much harm as good (if not more).
AI in Industrial Maintenance: Perfect Solution or Drunken Uncle?
During a recent interview, Rombouts identified GenAI’s ability to generate plausible sounding but incorrect information as one of the biggest risks of AI adoption. He likened this phenomenon to the drunken uncle at a family gathering.
Imagine sitting at a table where everyone is discussing investments. Among them is your uncle, who has had one too many drinks. He’s confidently offering stock market advice. Since you’re sitting down and having a face-to-face conversation with him, you can easily tell that he’s under the influence of alcohol. As a result, you question the validity of his advice.
Now imagine that you only read a transcript of his speech. His word choice and syntax convince you of his market knowledge and instill confidence that his recommendations are wise.
Businesses face a similar concern with Generative AI. It produces content that looks and sounds correct. However, users without the necessary expertise may struggle to discern whether it’s truly accurate or just well-articulated nonsense.
That’s a critical problem in industrial maintenance — incorrect guidance or instructions from GenAI could lead to mechanical failures or even fatal accidents.
Rombouts proposes testing an AI model’s ability to “detect BS.” The process is quite simple: He suggests purposely giving an AI model misinformation to see if it could separate fact from faulty data. He also recommends training maintenance professionals to spot unreliable AI output.
Think of these redundancies as a form of checks and balances. The model has been trained to sniff out incorrect information and flag it. Maintenance personnel are also trained in BS detection, allowing them to find and address any errors that make it past the model.
AI as a Tool, Not a Replacement
Artificial intelligence isn’t a magic pill that can solve all of your industrial maintenance headaches. It’s a powerful tool that’s already delivered huge advancements in preventive maintenance. However, its full potential will only be realized when trust, validation and human oversight are prioritized alongside adoption.
This article contributed by Tom Rombouts, Reliability Director