The Evolution of AI in Maintenance: From Expert Systems to Intelligent Agents

Ahmed Rezika, SimpleWays OU

Posted 8/6/2024

Maintenance teams have long been at the forefront of adopting intelligent systems, even before the term “artificial intelligence” (AI) became a buzzword. These systems, designed to mimic human intelligence and decision-making processes, have been integrated into various maintenance practices for decades, albeit under different names. 

For instance, expert systems encoded human expertise into a set of rules and heuristics. They have been employed in Computerized Maintenance Management Systems (CMMS) since the 1980s. [1] These systems assist in tasks such as scheduling preventive maintenance, diagnosing faults, and recommending maintenance actions. Essentially, they are acting as digital maintenance experts, or more commonly, your digital assistant. 

Growing Intelligence in Maintenance Assistance Tools

Maintenance teams have consistently embraced technological advancements, recognizing the importance of staying up to date on the latest tools and techniques. Smart alignment tools, for example, use laser or optical systems to precisely align rotating machinery. They are another example of intelligent systems that have been adopted by maintenance professionals. These tools often incorporate algorithms and decision-making capabilities to guide the alignment process and provide recommendations. Those recommendations come based on coupling type, equipment type, and system setup—effectively augmenting the expertise of maintenance technicians.

As the field of artificial intelligence continues to evolve, maintenance teams are poised to benefit from the latest advancements in generative AI. Generative AI models, capable of generating new content such as text, images, or even code, hold immense promise for maintenance applications. One potential use is in generating maintenance recommendations and reports. The end target is to generate tailored recommendations for maintenance tasks, repair procedures, or even predictive maintenance strategies. “Recommendations” is a key word in this target that we will explore later on. By training these models on historical maintenance data, expert knowledge, and best practices, they could prove to be a true asset. [2]

However, the adoption of generative AI in maintenance also raises concerns around many issues. One of them is: How will the maintenance team receive and accept this digital assistance? Another issue is the trust, interpretability, and the need for human oversight. The human supervision is crucial, as currently these models may exhibit biases or produce unexpected outputs. Nonetheless, as generative AI continues to mature, it presents an exciting opportunity for maintenance teams to further augment their decision-making processes and streamline operations.

AI in maintenance

What is AI? How does it work?

It is important for our maintenance teams to understand the foundations of the AI, how it works and its limitations. This way they will be well equipped for adopting some of these models in maintenance. Or at least they will be capable of dealing with it efficiently if adopted in the workplace.

Artificial Intelligence (AI) is a broad field that encompasses the study and development of intelligent systems and algorithms that can perform tasks typically requiring human-like intelligence. Examples of these tasks are problem-solving, decision-making, perception, reasoning, and even learning in advanced AI systems. This definition is a common one that captures the essence of AI and aligns our minds with the general understandings in the field.

At its core, AI aims to create systems that can acquire knowledge, process information, and make decisions. Then these systems take actions or generate a response based on that knowledge. AI systems are designed to mimic or augment human cognitive functions, such as learning, reasoning, and problem-solving. [3]

How does an AI system work?

In general, most AI systems work following a general process:

Building

Data Acquisition: AI systems require data to analyze, find patterns, recommend solutions, and learn. This data can come from various sources, such as sensors, databases, user inputs, or the internet for Generative AI. Just a note here, the data must be prepared to be useful. It is typically preprocessed and cleaned to prepare it for the learning process.

Feature Extraction: Relevant features or attributes are extracted from the raw data. These features represent the essential information that the AI system will use for learning and decision-making. Feature extraction in data is like summarizing a book. You take the raw data, which can be complex and lengthy, and identify the most important parts (features). These features are then used to make the data easier to analyze and understand by machines. That is crucial, especially for tasks like machine learning.

Model Selection and Training: An appropriate AI model or algorithm is selected based on the problem domain and the type of data. Further on we shall learn about the different models. The model is then trained on the preprocessed data and extracted features. During training, the model adjusts its internal parameters to learn patterns and relationships in the data. We will elaborate next on AI models.

Tuning

Model Evaluation and Validation: The trained model is evaluated on a separate test dataset to assess its performance and generalization ability. Evaluation metrics (e.g., accuracy, precision, recall) are used to measure the model’s effectiveness.

Deployment and Inference: If the model achieves satisfactory performance, it can be deployed in a production environment. When new data is provided to the deployed model, it can make predictions, decisions, or generate outputs based on the learned patterns. This process is called inference. Inference is like the AI model taking a test on unseen data. Basically, it’s the model putting its learned knowledge to work in the real world.

Continuous Learning and Adaptation – Updates: Many AI systems are designed to continuously learn and adapt as new data becomes available, enabling them to improve their performance over time. Otherwise we need to intend this to keep its output relevant.

In short, what is AI?

It’s important to note that while AI systems can exhibit intelligent behavior and perform tasks that would typically require human intelligence, they are ultimately complex algorithms and mathematical models designed to process and learn from data. The field of AI continues to evolve rapidly, with ongoing research and development aiming to create more advanced, adaptable, and reliable AI systems.

AI systems can employ various techniques and algorithms. An example of this is machine learning (supervised, unsupervised, and reinforcement learning). Another example is deep learning (neural networks) and natural language processing. The more handy examples are computer vision, expert systems, and optimization algorithms.

AI in maintenance

A Missing Term: The AI Agent

This term AI Agent is typically used in universities as a starting point for computer science students to help them comprehend AI and robotics concepts. However, when those AI models are introduced to users, we seldom listen to AI Agent. My first AI Agent that I programmed using Python could play tic-tac-toe. I followed it by another one that played Minesweeper.

So, let’s briefly understand the concept of an “AI Agent” even though it is being less explicitly mentioned nowadays. Because it was a central concept when studying AI, maintenance teams should know about it.

The idea of an AI agent refers to an entity that perceives its environment through sensors and acts upon that environment through actuators, with the goal of achieving some desired outcome or objective. An AI agent is essentially a decision-making system. It can observe, reason, learn, and take actions to maximize its chances of success in a given environment or task.

In the early days of AI research and education, the concept of an AI agent was heavily emphasized because it provided a useful framework for understanding and designing intelligent systems. It helped to conceptualize AI systems as rational agents that could perceive, reason, and act in an autonomous and intelligent manner, much like how humans and other intelligent beings operate in the world.

Why did the term AI Agent fade away?

As AI has advanced and evolved, the explicit mention of “AI agents” has become less prevalent, primarily for two reasons:

Abstraction and Specialization: As AI techniques and applications have become more specialized and diverse, researchers and practitioners often focus on specific subfields or tasks (e.g., computer vision, natural language processing, robotics) rather than discussing AI systems as general-purpose agents. The concept of an AI agent is an abstract and broad notion, whereas modern AI systems are often designed for specific, narrower tasks.

Integration and Ubiquity: AI technologies have become increasingly integrated into various systems, applications, and products, making the distinction between an AI agent and the overall system or application less noticed. Rather than explicitly referring to AI components as agents, the focus has shifted more towards the capabilities and functionalities added by AI, such as intelligent decision-making, automation, or adaptive learning.

While the term AI agent may not be as explicitly mentioned nowadays, the underlying principles and concepts remain relevant. Modern AI systems, whether they are explicitly labeled as agents or not, still possess the fundamental characteristics of perceiving their environment, reasoning based on that input, and taking actions or making decisions to achieve specific goals or objectives.

For example, a virtual assistant like Siri or Alexa can be considered an AI agent that perceives user input (speech or text), processes that input using natural language processing and other AI techniques, and then takes actions (providing information, executing commands, or controlling smart home devices) to fulfill the user’s requests. [4]

Do we still need the AI Agent? 

While the specific use of the term “AI agent” may have diminished, the core concepts and principles behind intelligent agents remain deeply ingrained in the design and development of modern AI systems, even if they are not always explicitly labeled as such. That’s why it was important to bring those concepts here to you.

Why Should Maintenance Teams be Aware of AI?

As new technologies are continually introduced in production environments to improve productivity, efficiency, and tolerances, maintenance teams have had to adapt and acquire knowledge to maintain these systems effectively.

Those advanced technologies are introduced in the production realm first, they then extend to the maintenance and other services. That’s why we need to prepare the maintenance teams to welcome AI technology. This will take us to deeper discussions of different AI models, how they can fit into maintenance, and if any of them are already used in the maintenance realm. Moreover, will the AI era replace the Industry 4.0? I will share insights about all of this in coming articles.


References:
  1. Terry Wireman, 2013, Successfully Utilizing CMMS/EAM Systems – Maintenance Strategy Series, Industrial Press, Inc.
  2. Salama Mohamed Almazrouei, Fikri Dweiri, Ridvan Aydin,  Abdalla Alnaqb, A review on the advancements and challenges of artifcial intelligence based models for predictive maintenance of water injection pumps in the oil and gas industry, Springer Link, Published: 21 December 2023, Volume 5, article number 391, (2023)
  3. Copeland, B.. “artificial intelligence.” Encyclopedia Britannica, Last updated  June 14, 2024. https://www.britannica.com/technology/artificial-intelligence.
  4. Wolfgang Ertel, 2017, Introduction to Artificial Intelligence, Second Edition, 2017, Springer

avt-img

Ahmed Rezika

Ahmed Rezika has over 25 years of hands-on experience in maintenance and project management. Ahmed is a Projects and Maintenance Manager with broad experience in industrial plants. He managed projects and applied different maintenance strategies and improvements tasks in different industries such as steel, cement, and food industries. He is certified as a PMP, MMP, and CMRP. Ahmed's goal is to create a better-managed value-adding working environment. Additionally, he established SimpleWays OU to contribute to a better-maintained world through training and coaching.

SimpleWays' vision is to support maintenance teams to add value to their organization. The team has implemented successful greenfield and upgrade projects over 25 years of experience in the steel and cement industries.