AI-Powered Maintenance: AI Models and Possible Use Cases
Ahmed Rezika, SimpeWays OU
Posted 9/4/2024
As we continue our exploration of AI-powered maintenance, it’s crucial to understand the landscape of available AI models and their potential applications in supporting maintenance activities. Before diving into specifics, let’s briefly revisit the concept of Artificial Intelligence (AI). At its core, AI refers to intelligent systems and algorithms capable of performing tasks that traditionally require human-like cognitive abilities. These include problem-solving, decision-making, perception, reasoning, and even creative creation in more advanced systems. By grasping the fundamentals of various AI models, maintenance professionals can better leverage these powerful tools to revolutionize their practices and improve operational efficiency. Read Part 1 here.
AI Models 101: Essential Knowledge for Modern Maintenance Professionals
What is an AI model?
An AI model is essentially the ‘brain’ of an AI system, containing algorithms and statistical techniques that enable it to process information and generate outputs. These models are trained on large datasets, allowing them to recognize patterns, make decisions, or solve problems in ways that mimic human intelligence. [1] In the context of maintenance, AI models can analyze sensor data, analyze historical equipment data and documents, predict equipment failures, optimize maintenance schedules, or even suggest repair procedures. The choice of AI model depends on the specific maintenance task-at-hand and the availability of plenty of data. However, you can train a machine learning model on limited data sets like 30 failure records or 20 production cycles but the accuracy of those model recommendations will be reduced. (The data landscape will be addressed in a later lecture.)
It is important for our maintenance teams to understand these foundations of the AI to be will be well equipped for adopting some of these models in maintenance.
Why it is important to categorize AI models?
Understanding the different types of AI models is crucial for maintenance leaders to effectively leverage benefiting from this technology. Categorizing AI models offers some key benefits:
Firstly, it helps in understanding the capabilities and limitations of each model type. Different models excel at specific tasks, and recognizing these strengths and weaknesses is essential for selecting the right tool for the job. Additionally, understanding limitations prevents unrealistic expectations and disappointments. Moreover, by understanding the various model types, maintenance leaders can quickly identify potential candidates for a specific problem, saving time and resources.
Secondly, clear categorization facilitates effective communication among data scientists, engineers, and maintenance leaders. Using standardized terms ensures everyone is on the same page, and it simplifies complex concepts for non-technical stakeholders. By organizing knowledge into distinct categories, maintenance teams can build upon existing expertise and identify focus areas where further research or development is needed.
Then, what are the popular AI models?
1. Rule-Based Systems
Rule-based systems operate on predefined rules and logic. They are ideal for tasks with clear and established procedures, such as troubleshooting common equipment issues. While effective for structured problems, their adaptability to new situations is limited.
Example in maintenance: Expert systems for troubleshooting common equipment issues, setup of testing equipment and processing maintenance tasks within a computerized system. Historically we used it in diagnostic systems, such as those for medical diagnosis. It is common in early AI implementations for tasks like tax preparation, and basic customer service bots.
2. Machine Learning (ML) Models
Machine learning models, on the other hand, learn patterns from data and make predictions or decisions without explicit programming i.e. without explicit rules. This versatility makes them suitable for a wide range of maintenance challenges.
For instance, maintenance can predict equipment failures based on historical data by identifying similar behavior in new data. ML algorithms can be as simple as linear regression or as complex as decision trees.
Subcategories of the Machine Learning Models:
- Supervised Learning: Models learn labeled data to map input data to output labels. (e.g., you record sensors data all the time and you record when machine is running and when it fails. Lastly, you ask the model, can you give an alert 1 day before failure?)
- Unsupervised Learning: Models find patterns in unlabeled data. [3] (e.g., identifying equipment groups with similar behavior as similarities in maintenance patterns.)
- Reinforcement Learning: Models learn by interacting with an environment through receiving feedback in the form of rewards or penalties. Then, it builds its own rules based on this feedback. (e.g., optimizing maintenance schedules based on positive KPI results from the planning dashboard)
3. Deep Learning Models
Deep Learning is a subset of machine learning that uses neural networks with multiple layers (hence the term “deep”) to learn from vast amounts of data.
Neural Networks are a specific type of algorithm inspired by the human brain’s structure. They are particularly adept at handling complex patterns and large datasets. [1] In an artificial neural network, cells, or nodes are connected. Each cell processes inputs and produces an output that is sent to other neurons. Data moves through the nodes, or cells, with each cell performing a different function. This process continues through multiple layers until the final output is produced.
Each connection between neurons has a weight associated with it. The incoming data from previous neurons is multiplied by these weights. These weighted values are then summed together.
All deep learning models are neural networks. However, not all neural networks are deep learning models. Not only Neural networks but also other simple algorithms are used in machine learning.
Example of usage in maintenance can be the image recognition for or during equipment inspection. Predicting a motor in an image based on being trained on many images labeled as motor
4. Large Language Models (LLMs)
Large Language Models (LLMs) are indeed a distinct category of AI models. They are specifically designed to process and generate human-like text. While they have shown remarkable capabilities in various fields, their direct application in maintenance might be more focused on tasks like:
Generating maintenance reports: Automating the creation of detailed reports based on collected data.
Documenting maintenance procedures: Creating clear and concise step-by-step guides.
Analyzing maintenance logs: Extracting valuable insights from textual data.
While LLMs offer potential [4], their practical application in maintenance is still an emerging area that has some risks and will be discussed in the coming article.
A Handy Guide to AI Models
Now we shall have a balanced overview of each model type, including their definitions, operations, applications, advantages, and limitations.
Just we need to keep in mind that in the maintenance realm we don’t target creating or coding an AI model from scratch. Our efforts are directed towards being prepared to onboard an AI-based systems in our daily work, teach it and make the best use out of it.
Rule-Based AI:
– Definition: Rule-based AI, also known as expert systems, operates on a predefined set of rules and logic dictated by human experts. These systems make decisions by following these explicit instructions.
– Operation: Typically uses something like “if-then” statements to process inputs and produce outputs. The system’s knowledge base(data) and inference engine (applies logical rules to a knowledge base to reach conclusions) work together to apply preprogrammed rules to the given data.
– Applications: Historically used in diagnostic systems, such as those for medical diagnosis[4] or machinery troubleshooting. Common in early AI implementations for tasks like tax preparation, and basic customer service bots.
– Advantages: Highly transparent and interpretable, making it easy to understand and trust the decisions made by the system. They are also relatively straightforward to design and implement for well-defined problems.
– Limitations: Limited flexibility and adaptability, as they can’t learn from new data or handle unforeseen scenarios. Maintenance of the rule set can become complex and cumbersome as the system grows to include more options i.e. more decision points and forks.
Supervised Learning Models:
– Definition: Supervised learning models are trained on labeled data, where both input and desired output are provided.
– Operation: These models learn to map inputs to outputs based on example pairs. They adjust their parameters to minimize the difference between their predictions and the actual labeled outputs.
– Applications: Commonly used for classification (e.g., categorizing equipment faults) and regression (e.g., predicting remaining useful life of machinery) tasks. Regression analysis is essentially about finding a mathematical equation that best fits a set of data points.
– Advantages: Can make accurate predictions on new, unseen data if trained properly. Useful for problems where the relationship between input and output is complex but examples are available.
– Limitations: Requires large amounts of labeled data (Solved problems), which can be expensive or time-consuming to obtain. May struggle with scenarios very different from its training data.
Unsupervised Learning Models:
– Definition: Unsupervised learning models work with unlabeled data, trying to find patterns or structures within the data itself.
– Operation: These models identify similarities or differences in the data, often grouping similar data points together through clustering for example or reducing data complexity.
– Applications: Useful for anomaly detection in sensor data, grouping similar maintenance issues, or discovering hidden patterns in equipment behavior.
– Advantages: Can uncover hidden patterns in data that humans might not notice. Doesn’t require labeled data, which is often more abundant and easier to collect.
– Limitations: Results can be more difficult to interpret or validate compared to supervised learning. The patterns found may not always align with what’s most useful for the specific task.
Reinforcement Learning Models:
– Definition: Reinforcement learning models learn by interacting with an environment, receiving feedback in the form of rewards or penalties.
– Operation: The model (agent) takes actions in an environment and learns which actions lead to the best outcomes over time, optimizing its behavior to maximize cumulative rewards.
– Applications: Can be used for optimizing maintenance schedules, energy management in facilities, or autonomous robotic maintenance tasks.
– Advantages: Can learn complex strategies and adapt to changing environments. Well-suited for sequential decision-making problems.
– Limitations: Can be challenging to define appropriate reward structures. Training can be time-consuming and may require many iterations. The learned strategies may not always be interpretable or explainable.
Deep Learning Models
– Definition: Deep learning models are a subset of machine learning that utilize artificial neural networks with multiple layers to learn complex patterns from large amounts of data.
– Operation: Deep learning models process data in stages, passing it through multiple interconnected layers of artificial neurons. Each layer extracts different features from the data, gradually building up complex representations. The model learns by adjusting the weights and biases of these connections through a process called backpropagation.
– Applications: Widely used in image and speech recognition, natural language processing, and increasingly in maintenance for tasks like equipment image analysis, predictive maintenance based on complex sensor data patterns, and anomaly detection.
– Advantages: Capable of learning highly complex patterns from large datasets. Often outperform traditional machine learning methods in tasks involving unstructured data.
– Limitations: Require significant computational resources and large amounts of data. Black-box nature makes it difficult to interpret how decisions are made.
Large Language Models (LLMs)
– Definition: Large Language Models (LLMs) are a type of artificial intelligence that can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
– Operation: LLMs are trained on massive amounts of text data to learn patterns of language. They can generate text, translate languages, write different kinds of creative content, and answer your questions in an informative way.
– Applications: Potential applications in maintenance include generating maintenance reports, creating documentation, and analyzing maintenance logs for insights.
– Advantages: Can process and generate human-like text, enabling new forms of interaction and automation.
– Limitations: May generate incorrect or misleading information. Require significant computational resources and large amounts of data to train.
Navigating the AI Landscape in Maintenance
As we’ve explored, the world of AI models offers a rich array of tools that can revolutionize maintenance practices. From rule-based systems that excel at structured problem-solving to advanced deep learning models capable of uncovering complex patterns in vast datasets, each type of AI model brings unique strengths to the maintenance field.
Understanding these models is not just about keeping up with technology—it’s about unlocking new possibilities for efficiency, predictive capabilities, and decision-making in maintenance operations. As maintenance professionals, familiarizing ourselves with these AI models prepares us to a brighter tomorrow.
LLMs -publically called GPTs- became popular in the last couple of years in such a way that some people match this to the start of the internet era. While it seems the easiest model to use and most commercially available model, it needs a cautious mindset to avoid its traps. Let’s go to this landscape of the AI in our coming lecture.
References
- Sara Brown “Machine learning, explained” MIT Management Sloan School, Last updated April 21, 2021. https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained
- IBM, Supervised versus unsupervised learning: What’s the difference? 12 March 2021,
- Awasthi, Raghav & Mishra, Shreya & Grasfield, Rachel & Maslinski, Julia & Mahapatra, Dwarikanath & Cywinski, Jacek & Khanna, Ashish & Maheshwari, Kamal & Dave, Chintan & Khare, Avneesh & Papay, Francis & Mathur, Piyush. (2024). Artificial Intelligence in Healthcare: 2023 Year in Review. 10.1101/2024.02.28.24303482.
- Alhussein Fawzi and Bernardino Romera Paredes, Dec 2023, FunSearch: Making new discoveries in mathematical sciences using Large Language Models, Google DeepMind.
Related Articles
Keys for Effective Troubleshooting
Analyzing Semiconductor Failure
The Lubrication Requirements of Couplings
What is the True Downtime Cost (TDC)?
Improvement: What Comes First?
Use P-F Intervals to Map, Avert Failures
