From Descriptive to Predictive: Shaping the Future of Equipment Maintenance Through Data Science
The manufacturing industry is on the verge of a data‑driven revolution. For years, descriptive analytics has been telling us what just happened and why, however, in this era where technology advances unbridled, staying in such type of analysis is no longer a competitive advantage over other companies. That is why manufacturing companies are betting towards the future: using the predictive aspects of big data to monitor their operations and foretelling any breakdown of their equipment or quality standards.
Predictive analytics isn’t a new discipline; however, it used to be a very expensive field due to the highly specialized equipment used to carry out the forecast. Companies invested in these analyses for very specific cases where large amounts of money were in between. However, after some years, we finally begin to see how predictiveness becomes more common in manufacturers daily operations thanks to the creation of new tools and software accessible to the public. On the other hand, as the Internet of Things continues to mature, manufacturers are gathering more and more data automatically.
Manufacturers know the importance of relying heavily on data for decisions making, but many of them are just beginning to discover the great impact that can be reached out through maximizing the value of their tools and machines. Now a days, IoT-connected machines can measure, register, and transmit real-time data, enabling manufacturers to uncover insights that can improve performance, create more efficient production runs, and maximize uptime for machines (Bhaskar A., 2022).
One of the manufacturing areas that will have a great impact thanks to predictive analytics is equipment maintenance. With increasingly sophisticated machinery embedded with IoT sensors, now a days maintainers have plenty of variables that can be analyzed for decision making such as temperature, vibrations, and others, but also non-technical variables such as operators’ performance: who’s working on which machine, how long of a shift they’ve worked, what tools are being used in the press, etc.
“A study conducted by Honeywell and KRC found that effective use of big data analytics in a manufacturing company can reduce breakdowns by up to 26 percent and extends equipment lifetime by 20 percent” (Consoli R., 2018).
Collection of data is the jumping off point for predictive maintenance, and it will only work if is time-series values, meaning it’s collected at specific, discrete times. Theoretically, the data collection process seems simple, but the reality is that it is a stage with a high probability of presenting errors and if it is not well executed, none of our next steps will be reliable; “garbage in, garbage out”.
All technical and human factors data can be analyzed altogether to create mathematical models. “Regression is a powerful and well-known forecasting methodology to predict future fault events based on trends found on time series” (Dotis-Georgiou, A., 2021). However, this is not the only mathematical strategy, there are several methodologies that can be combined to improve the performance of prognostics systems.
Although, data-driven predictiveness seems like a top tier, the reality is that now a days predictivity is just one step to reach a higher level: prescriptive machinery. Machine Learning enables self-optimizing systems that take autonomous action through self-learning and self-steering algorithms, with input from historical and real-time data (Burns E., 2021). Obviously, this is not an overnight task; having a predictive and autonomous process is a complex job that can take years, from data collection and verification, modeling, and the machine’s learning process.
The P-F curve allow us to visualize the behavior of different types of maintenance against time and the equipment’s health. As predictive maintenance uses sensors and data to detect trends in the health of a system, it allows it to perceive the deterioration of a machine way earlier than condition-based, giving more time to schedule maintenance at a convenient time, before failure occurs, but without wasting resources on unnecessary maintenance. The new era of maintenance is called reactive or prescriptive, and results from the usage of advanced data analytics and machine learning to determine the root cause of a potential failures so specific corrective action can be prescribed automatically. Also, potential failure can be identified even earlier than in predictive maintenance, which makes fixing the problem easier and less expensive (Hanly S., 2020).
Without a doubt, data science came to revolutionize the world; manufacturing is one of the industries that must accelerate the use of predictiveness in all its processes. However, it should be emphasized that forecasting not only relies on the collection and analysis of large amounts of data, but also on changing the mindset of the organization. Implementing predictive maintenance requires a significant investment in money, personnel, and education; specialized equipment and a trained team that leads the new processes and procedures are necessary. While these initial investments might seem intimidating to an organization, predictive maintenance’s return on investment far outweighs any upfront costs.
References:
- Bhaskar, A. (October 20, 2022). How IoT Is Playing A Key Role In Production Uptime [Forbes]. Website: https://www.forbes.com/sites/forbestechcouncil/2022/10/20/how-iot-is-playing-a-key-role-in-production-uptime/?sh=90b92661f200
- Consoli, R. (May 7, 2018). Using Big Data Analytics To Improve Production [Industrial Media]. Website: https://www.manufacturing.net/operations/article/13228439/using-big-data-analytics-to-improve-production
- Dotis-Georgiou, A. (July 22, 2021). An introduction to time series forecasting [InfoWorld]. Website: https://www.infoworld.com/article/3622246/an-introduction-to-time-series-forecasting.html
- Burns, E. (March 25, 2021). Machine Learning [TechTarget]. Website: https://www.techtarget.com/searchenterpriseai/definition/machine-learning-ML
- Hanly, S. (November 24, 2020). Differences Between Condition-Based, Predictive, and Prescriptive Maintenance [Endaq]. Website: https://blog.endaq.com/differences-between-condition-based-predictive-and-prescriptive-maintenance
