Predictronics had a rewarding experience at this year’s 11th Annual Conference of the Prognostics and Health Management Society in Scottsdale, Arizona.

At the conference, Predictronics CTO Dr. David Siegel served as a panel co-chair and presented at the Measurement and Evaluation for PHM in Manufacturing (ME4PHM) workshop. Dr. Siegel’s presentation focused on the development and capabilities of PHM technologies for manufacturing, as well as the challenges and best practices in integration and deployment.

The goal of this workshop was to provide an opportunity for industry professionals, technology providers, government officials and academics to share their expertise and debate critical topics in PHM for manufacturing. Topics discussed include measuring, verifying and validating PHM technologies; assessing PHM technologies for small manufacturing versus large manufacturing; measuring and evaluating methods in PHM, such as value determination and maintenance techniques; and researching and exploring the latest in PHM technologies, such as Industrial AI for PHM.

Dr. Siegel shares his top four takeaways from the Measurement and Evaluation for PHM in Manufacturing workshop:

1. Similarities and Differences Between Large and Small Manufacturers

Two separate discussions took place at the workshop extrapolating the similarities and differences in PHM technologies for large and small manufacturers. In most cases, both need robust PHM solutions that create real-world value and both utilize the same types of equipment, such as CNC machines and robots.

For small manufacturers, there are far less obstacles to overcome when starting a PHM initiative, as well as less IT considerations. However, large manufacturers have greater internal resources and budget allocations for investing in the development and adoption of PHM solutions.

At Predictronics, most of our customers are large manufacturers, so the workshop was an excellent opportunity to learn about the challenges and needs of small manufacturers and determine how we can apply this understanding to our customer solutions in the future.

2. Emerging Technologies for Prognostics and Health Management

Deep learning was a popular topic explored during the panel. Recently, researchers in the deep learning community have started to delve into more challenging aspects of the application of deep learning in PHM, including how to incorporate domain knowledge and physical models into deep learning models for anomaly detection and diagnosis. In addition to deep learning for hybrid models, transfer learning could be another compelling method to examine, as the training data for each asset or asset variant is limited.

New research was presented at the workshop highlighting explainable AI (XAI). Explainable AI provides contribution values and other outputs from analysis models to give a better understanding of how the AI makes its decisions and predictions.

With the use of more complicated machine learning models and the necessity of human knowledge and experience, the topic of XAI has only continued to grow in popularity. Predictronics currently provides explainable AI solutions through our team’s domain knowledge, predictive modeling expertise and experience in software development and deployment.

3. Identifying the Value in PHM Solutions

One universal theme examined during the workshop was the relationship of various PHM accuracy metrics to actual return on investment (ROI).

Performance metrics, such as ROC curves, early detection frequency indicators, false alarm rates and confusion matrices can be calculated, but these do not provide a direct correlation to business-related metrics. Ideally, if the business-related metric can be defined in relationship to the analysis accuracy metric, this could influence the tuning of the analysis model and could create a bias in the sensitivity of the model.

Predictronics proves ROI by implementing more rigorous methodologies and generating numerous example case studies to demonstrate a link between business objectives and analysis performance metrics.

4. The Maintenance of the PHM Solution

One key challenge with PHM technologies and solutions is maintaining them, which includes the adoption and fine-tuning of analysis models. Many factors can cause a shift in data over time, including maintenance actions, process drift, new operating regimes and new failure modes.

Businesses can ensure the accuracy of their predictions and add significant value to their PHM solution by incorporating models that retrain with new baselines, adapt to drift and learn new fault patterns.

Predictronics’ template-based approach provides a base solution for many key industry assets and applications that can be quickly deployed and customized for customers. However, tasks such as fine-tuning, maintaining and retraining models, offered through Predictronics consulting services, still require manual intervention. Providing understandable and efficient tools for conducting these tasks or automating them entirely in the future would lead to more sustainable PHM solutions for customers.

PHM 2019 was a great success overall, fostering a dialogue through its many panel sessions, short courses and workshops. The event gave attendees a better understanding of the standards and methods guiding PHM technologies, as well as the impact of these technologies on other disciplines and real-world applications.