Artificial Intelligence is beginning to challenge a fundamental assumption of maintenance – that the maintenance engineer is the primary source of knowledge about the condition of an asset.

When machines or software can predict failures, diagnose problems and recommend actions, what remains for the maintenance engineer in terms of engineering judgment?

Evolution of maintenance

In recent years, big data became a big focus in industrial maintenance. Organisations invested heavily in data collection systems and computerised maintenance management systems (CMMS), gathering information from vibration monitoring, process control equipment, and many other sources.

Prior to that era maintenance had already progressed from reactive 'run-to-failure' strategies to planned preventive maintenance, and then to condition-based and predictive approaches. Industry 4.0 accelerated this transformation by connecting machines, control systems and business information systems and allowing data to be exchanged in real time. Digital technologies changed how assets were monitored, maintained, and managed throughout their life cycle.

However, more data did not in itself produce better maintenance decisions. The challenge was identifying what useful information could be derived from the data and converting it into actions. Data alone does not reduce risk, prevent failures, or improve reliability. The value comes from converting it into reliable information, useful insight, and an appropriate decision.

AI is now emerging as a powerful new capability within this digital maintenance environment. When used properly, it can help organisations extract greater value from the information already available.

The changing role of the maintenance engineer

Increasing equipment complexity, ageing assets, higher production expectations, and shortages of experienced personnel are driving a different approach to maintenance and reliability management. AI can be used to recognise fault patterns in equipment behaviour that may be difficult to detect through conventional analysis. This capability could help maintenance teams identify developing faults and determine where intervention may be justified. Predictive analysis of this type is also likely to support maintenance planning, scheduling, and resource allocation.

The potential benefits are considerable if the system is applied correctly. An AI system may, for example, identify a deteriorating bearing condition weeks before failure. But identifying a developing fault is only the beginning of the engineering decision.

While AI identifies patterns, relationships and anomalies within the data and models available to it, it does not understand the wider operating context in the same way as an experienced maintenance engineer will. A technically impressive prediction can therefore still lead to a poor decision if that context is misunderstood.

Taking the example mentioned where an AI system identifies a bearing as having a high probability of failure within the next few weeks, the maintenance engineer must consider many variables such as the credibility of the prediction, the consequences of failure, operating conditions, production requirements, safety implications, spare availability, and the opportunity cost of intervention.

The prediction is not the decision; it is the response to that prediction that defines the outcome. This is where solid engineering judgment and understanding of context become critical. It can be expected that engineering judgment will remain critical in areas such as:

• Recognising when data is incomplete;
• Understanding physical failure mechanisms;
• Assessing uncertainty in the analysis;
• Challenging a plausible diagnosis;
• Understanding operating context;
• Balancing safety, reliability, cost, and production;
• Determining the appropriate circumstances for intervention;
• Recognising when a model is operating outside its competence;
• Taking responsibility for the resulting decision.

Equipment design is also changing in response to AI capability which will add to the complexity of information. We can expect that machines will increasingly be designed not only to perform their intended function but also to provide continuous information about their own health and performance. This will strengthen the connection between design, operation, and maintenance.

AI also has the potential to connect decisions across the asset life cycle. Information generated during design, commissioning and operation can increasingly be carried forward to improve future specifications, maintenance strategies, refurbishment, and replacement decisions. Actual operating experience today could therefore influence the design and management of future assets.

Moving forward the maintenance engineer will consequently require a broader skill set. Traditional engineering knowledge will remain essential but must be complemented by an understanding of data analytics, automation, cybersecurity, and digital asset management. RCM, FMEA and RCA will remain fundamental, but AI can accelerate them by analysing large datasets and identifying trends.

Risks that cannot be ignored

Every technological advance creates opportunities and risks. AI is no exception. Successful implementation depends on people, processes, and engineering discipline as much as technology.

The experience of CMMS provides a useful warning. As the technology became increasingly available during the 1980s and 1990s, many organisations adopted it without necessarily exploiting its full potential. Technology implementation alone does not guarantee value.

Data, processes, people, leadership, and engineering capability must work together if the technology is to deliver value.

SOLAS identifies maintenance, engineering, and services technicians, including electrical and mechanical maintenance roles, among difficult-to-fill occupations¹. Organisations are also facing the loss of experienced personnel while modern facilities require expertise in reliability engineering, automation, digital technologies, and data analytics. Long-term success will depend on workforce development, structured knowledge transfer, collaboration with educational institutions and continuous engineer development.

Data quality – can we trust it?

AI is only as reliable as the information and context available to it. Poorly calibrated sensors, incomplete maintenance histories, inaccurate work orders, and inconsistent failure coding can produce misleading recommendations. The principle of 'garbage in, garbage out' remains highly relevant: AI can only produce reliable insights when the underlying asset, operational and maintenance data is accurate, complete, and consistent.

The integrity and quality of operational data are fundamental to effective AI-based maintenance. Organisations must have confidence that data remains accurate as equipment, operating conditions and maintenance practices change. AI models must also be regularly reviewed to ensure that their recommendations remain relevant.

If maintenance engineers cannot trust the data, they cannot trust the recommendation derived from it.

The scale of the problem becomes greater when AI is applied across an entire asset population. Poor data could generate misleading recommendations for hundreds or thousands of assets.

Effective AI therefore depends on high-quality asset information, including:

• Well-structured asset hierarchy;
• Consistent failure coding;
• Accurate maintenance history;
• Reliable sensor data;
• High-integrity process information.

Protecting engineering capability

There is another, less obvious risk. As AI assumes a greater role in fault diagnosis and maintenance planning, maintenance engineers may gradually lose the practical troubleshooting skills traditionally developed through experience.

GPS did not eliminate navigation; it simply reduced the number of people who could read a map with confidence. AI could have a similar effect on troubleshooting skills.

A young engineer who has always relied on AI to identify bearing, pump or gearbox faults may become particularly good at responding to recommendations without developing the ability to diagnose unfamiliar failures from first principles. The question is therefore not simply whether AI can diagnose a fault, but whether maintenance engineers will still be capable of operating effectively when AI is wrong, unavailable or outside its area of competence.

Engineers must continue to investigate failures, challenge assumptions, verify recommendations and establish underlying causes. These skills develop through experience, exposure to unfamiliar problems and learning from incorrect diagnoses. AI should support that development rather than replace it.

Accountability and decision ownership

For safety-critical engineering decisions, organisations must retain clearly defined human accountability, regardless of how capable AI becomes. AI can recommend an action, but it cannot accept legal, engineering, or ethical accountability for the consequences.

Consider a future AI maintenance system recommending 'Replace pump PU04A04 drive-end bearing within 10 days'. The recommendation is not the end of the process. Somebody must establish whether the diagnosis is valid, whether replacement is necessary, whether the plant is safe to operate, whether the correct spare is available and whether the proposed intervention is appropriate. Decision support does not equal decision ownership.

What should organisations do now?

Organisations should prepare by improving data quality, investing in workforce capability, strengthening cybersecurity, and embedding AI within established asset management processes rather than treating it as a standalone technology.

When used to the full, AI can remove much of the routine work involved in reviewing alarms, maintenance histories, and condition-monitoring data. This creates an opportunity for engineers to spend more time on failure investigation, risk assessment, asset strategy, optimisation, and complex decision-making.

The organisations that gain the greatest value from AI will not necessarily be those that automate the most decisions. They will be those that combine intelligent technology with skilled people, reliable data, and sound engineering processes.

Conclusion

Engineering has never been simply about finding answers; it has also been about asking the right questions.

AI can provide powerful insights and may become better than humans at recognising patterns within large and complex datasets. But engineering is not simply pattern recognition. It involves deciding what those patterns mean, what should be done about them, what risks are acceptable and who is accountable for the decision.

AI could therefore make engineering judgment more important while simultaneously making it harder for people to develop that judgment. The challenge for the maintenance engineer is to adopt AI without losing the practical experience, critical thinking and responsibility that underpin good engineering decisions.

The engineer of the future will not be defined by access to information, but by the ability to think critically, challenge technology when necessary and apply engineering judgment with confidence. 

Author: John Coleman, Retired, BSc. (Hons) MEng MIEI, MIAM, MI Ref Eng is a former employee of Rusal’s Aughinish Alumina Refinery in Co Limerick. During his time there he worked in the area of contractor management and maintenance management. He is a former chairman of MEETA Asset Management which is a sector in Engineers Ireland aimed at promoting maintenance. A former chair of Engineers Ireland Thomond Region he is also a member of the Institute of Refractories Engineers and The Institute of Asset Managers (IAM). He sits on the Engineers Ireland Council and Executive boards. His professional experience spans more than 30 years in high volume manufacturing including management positions in planning, maintenance, operations and reliability.