Mechanical Reliability in the Chemical Industry: Challenges and Successes in Predictive Maintenance Modeling

To be successful in the chemical industry, safety is first and foremost and reliability is key. Unexpected equipment failures can mean lost production, higher emissions from abnormal conditions and process swings, and more risk to those working in the facility. Over the past decades, maintenance strategies have evolved in the chemical industry to minimize unexpected equipment failures along with the associated cost and production losses. Historically this is done by understanding mechanical failure mechanisms, creating preventative maintenance programs based on the mean time between failures, and monitoring key performance indicators.

There exists a wealth of data in the chemical industry in process instrumentation and historical equipment records, and advances in digitization and data utilization are occurring in the age of Industry 4.0. Data mining, increased computing power, and user-friendly implementations of machine learning models enable early detection of impending failures, in some cases with significantly improved lead times compared to condition based monitoring systems. However, challenges exist in modeling these sparse datasets (few failures to train a model on), on equipment that is not like-for-like across an enterprise, and where the end-users do not have data science expertise.

In this talk we will discuss the improvements in reliability that machine/deep learning models can provide along with the challenges the chemical industry faces in applying these methodologies. Dow Chemical is undergoing an evaluation of many current technologies on the market, and we will share key elements in our assessment of the capabilities and the fit for our industry. There are challenges with the “big data” we have available, but there are successes to share as well, ultimately leading to more reliable, sustainable, and safe chemical plants.