Morgan Vawter, Chief Analytics Director, Caterpillar
Industrial firms are betting big on advanced analytics and big data projects with expectations for high return on investment and short time to value. The explosion in growth of sensors, low-cost cloud data platforms, and easy-to-use visualization tools enable the connection and optimization of equipment, people, jobsites and infrastructure. However, getting the most out of your investment requires the proper ratio of technology, people, and processes. Combining data science with domain expertise brings context to the information, connection to KPIs (what’s this?), and enables a compelling narrative about the potential impacts. Informed decisions require relevant, timely insights, and clear decision owners. Taking action to implement decisions requires business processes, process owners and business process monitoring.
Let’s take a closer look at where data science meets smart iron at Caterpillar Inc, the world’s leading manufacturer of construction and mining equipment, diesel and natural gas engines, industrial gas turbines and diesel-electric locomotives. With over half a million connected assets, Caterpillar has the largest connected industrial fleet in the world.
Just as industrial manufacturing firms have invested in sensors and process controls for assembly lines, so have equipment OEMs by building them into systems, assemblies, and even individual parts. Construction and mining machines are analogous to mobile factories and when those factories aren’t producing, our customers aren’t making money.
With asset health in mind, our customers typically evolve from reactive to predictive analytics leveraging machine learning to avoid downtime and more effectively use planned downtime.
Vision is the primary and fastest way to process information
Achieving these potential benefits, though, requires more than just good data science. In many cases, it helps to have a solid understanding of how the equipment is designed, how it works, how it’s used, and how it fails to really connect the dots and understand all that the data is saying. For example, a number of algorithms might be able to detect that exhaust temperature is abnormal, but it might take an understanding of the engine to tell if the cause is a faulty injector or an after cooler leak. Data science can look for correlations between variables, but domain expertise can help focus the efforts on variables most likely to be related. It’s this symbiosis of data science and domain expertise that often delivers value beyond the sum of its parts.
While the analytics models and tools might be well understood by any data scientist, it’s equally important to have knowledge around the engine application, operating environment, configuration and upstream/downstream components. As an example, a mining customer spent $650,000 and lost over 900 hours of production as a result of unplanned downtime. By using our predictive analytics models, the customer would have foreseen the issue and proactively planned for the repair with a cost of $12,000 and 24 hours downtime.
With the critical insights and recommendations in hand, the next step is to effectively communicate. Visualization is the key to understand the data as analysts work with it, and communicate the results to stakeholders. Effective visualizations should encompass more than a static story, by incorporating controls for the end-users to manipulate and interact with the representation. Almost 50 percent of the brain’s resources are dedicated to vision and over two-thirds of the body’s sense receptors are in the eye, therefore, vision is the primary and fastest way to process information.
Industrial companies are increasingly challenged to get the most out of our investments in big data, and the best path to maximize ROI is through agile development, which utilizes short sprints to deliver incremental wins. Consider using data-driven rapid improvement workshops and continuous improvement initiatives to connect your data scientists with technical experts and business stakeholders who will implement changes based on the analytics driven insights. Building small, project-focused applications, and solutions on top of a strong data model as a foundation, allows teams can move fast and deliver results quickly that meet diverse needs of the business. Delivering on the promise of big data is challenging, so it’s important to regularly reflect on the tactics and outcomes of each phase. Iterate, pivot, operationalize, and communicate wins within the enterprise before moving on to the next patch of incremental value.