AI for asset management and predictive maintenance

Introducing AI's role in asset management and predictive maintenance, including the activities and focus areas of the Institute for Manufacturing.

Explaining machine learning as a method for computer programs to improve performance through experience, without explicit programming. Differentiating AI and machine learning.

Discussing examples of AI applications in maintenance, decision-making support, and predictive maintenance in manufacturing contexts.

Explaining supervised learning, unsupervised learning, and reinforcement learning with examples such as number sequence logic and object categorization.

Detailing traditional data sources in manufacturing, the advent of IoT for vast data collection, and the impact on predictive maintenance.

Explaining how predictive maintenance uses data from condition monitoring to predict and optimize maintenance activities in manufacturing settings.

Discussion on the approach to analysis when labeled datasets are not available, emphasizing the importance of identifying the problem statement to determine the suitable machine learning algorithm.

Overview of the timeline for the widespread adoption of predictive maintenance in the industry, highlighting challenges such as the need for common data standards and cost barriers for smaller companies.

Explanation of the time-consuming process of creating and training models, focusing on data collection, preprocessing, and model development for different algorithms.

Addressing the change management aspect in integrating machine learning teams with conventional maintenance in companies, discussing the cultural shift and trust in machine learning algorithms.

Exploration of using external databases to support companies in developing predictive maintenance models, mentioning the availability of manufacturing datasets for training purposes.

Recommendation to prioritize condition-based maintenance before moving to predictive maintenance, emphasizing the benefits of data collection and identification of maintenance patterns.

Strategies for handling imbalanced datasets, including oversampling and using physics-based models or expert knowledge to generate synthetic data for model training.

Discussion on the potential of using condition-based and predictive maintenance for robots, mentioning challenges in obtaining data from robots for maintenance purposes.