While data science uses enterprise and social media data to optimize consumer engagement and experience, industrial data science heavily relies on time series data collected by machines through connected assets and sensors. Time series data is a sequence of measurements captured over time and stored in chronological order. Often, the time series recordings of the machine signals are not enough to drive a meaningful business outcome, so these data points have to be combined with the physical operating model of the machine.

In this article, we will use an aircraft engine to explain how industrial data science can be applied to the Internet of Things such as a passenger aircraft operated by airlines.

industrial1Figure 1: Different Flight Stages of Aircraft (Source: IATA)

To understand the physical operating model of the aircraft engine, let us look at how the engine operates for takeoff, climb, and cruise in the air, before it descends and lands at the destination. The combustion in the jet engine uses about 50 parts of air mixed with 1 part of aviation fuel, which results in exhaust gases from the engine reaching temperatures of 700 to 800 C. This is problematic because the engine’s metal parts start to degrade when it is exposed to high temperatures for a prolonged period of time. To record the heat exposure levels, the engine has multiple sensors that measure the temperature of the exhaust gases in various places, along with pressure, vibration, and other parameters. Figure 2 plots the temperature profile as the engine is turned on, taxis, takes off, climbs, and reaches the flying altitude. Such sensor data is stored in Flight Data Recorder (FDR) in the aircraft. A data scientist can generate such plots using the time series data obtained from FDR after the aircraft lands—for one flight, all flights of a specific engine, or an entire fleet of such engines. However, to drive a business outcome—such as predictive maintenance for the engine in order to reduce unplanned downtime of the aircraft—requires a combination of math-based and physics-based models.

industrial2Figure 2: An Aircraft Jet Engine's EGT (Exhaust Gas Temperature) Plot

In the math-based approach, the historical data plots over several flights produces a pattern as shown in Figure 2. If a new flight data when plotted shows a large deviation, then an alert will be generated. Data scientists are good at detecting these types of deviations. However, they struggle to figure out the impact of this deviation to the condition of the engine and the impact to future flight operations. This is where industrial data science requires the product engineers—in this case, the aviation engineers—to work with the data scientist and assign meaning to the observed data by using the knowledge of how the product is designed to operate (i.e., the physics-based model).

When the aircraft is on the gate, it is at room temperature and when the engine is turned on, the temperature starts to rise. As it taxis and takes off, the EGT rises. In the climb phase, the EGT rises very quickly because the engine is working hard to gain altitude. The faster the aircraft has to climb to reach the cruise altitude, the higher the EGT will be. If the rate of climb is fast, the aircraft reaches the cruising altitude sooner, which is the most fuel efficient stage and has the least operating stress on the engine. On the flip side, the higher the heat exposure of the engine parts to temperature above threshold, the faster the parts will degrade over flight cycles. This means that the engine must visit the shop for maintenance more often so these heat-exposed parts can be replaced before failure. By plotting the EGT for a given engine and its historical flights, you can find the cumulative heat exposure above the temperature threshold and predict the right time to schedule the engine for maintenance. This is an example of applying industrial data science to a known problem.

Industrial data science has made way for the digital twin, a virtual representation of a physical asset, object, or process. In the example above, the digital twin is the virtual representation of the aircraft and may include several digital twins of major parts like the engines and avionics. These digital twins allow the industrial data science to track the health of the physical assets by collecting the sensor and other contextual data leading to reduced downtime via predictive maintenance and operations optimization. The detailed explanation of digital twin is outside the scope of this post and will be covered in a future post.

Shyam Nath
Shyam Nath

Shyam Nath is the author of the book “Architecting the Industrial Internet.” He is an Enterprise Cloud Architect at Oracle and focuses on IoT, Blockchain, and Advanced Analytics. Prior to Oracle, he worked for GE, IBM, Deloitte, and Halliburton. He has an MBA and MS in Computer Science.