For past couple of years, Engineering Analytics is gaining popularity within the product manufactures & asset-heavy organizations working in the operations space. The sectors like automotive, aerospace, medical devices, mining, oil & gas are investing heavily to make their products and processes smarter by incorporating multitude of sensors and connectivity solutions. Engineering Analytics is considered as one of the main avenues to monetize this investment in IoT/Industry 4.0 related solutions.
Owing to lack of a standard definition for engineering analytics, the scope of this topic is open and various organizations interpret it differently based on their own context. For us, Engineering Analytics is an approach where we make use of the data coming from engineering design, manufacturing, and operations to generate business value in terms of cost savings and additional revenue generation.
The typical data sources are engineering IT systems (PLM, MES, ERP) and machine data (PLCs, sensors, repair logs, error codes etc). Some of our popular service offers under Engineering Analytics include:
- Product Launch Analytics,
- Predictive Maintenance,
- Warranty Analytics,
- Manufacturing Cost Optimization,
- Factory Analytics,
- Asset Operations Analytics
We believe that executing engineering analytics projects needs a different approach altogether when compared to the standard non-engineering analytics work. All engineering processes and operations are governed by underlying laws of physics & mechanics. In the ideal world, the outcome of any engineering process can be represented using a set of closed-form equations. When it comes to the real world, it brings uncertainties (variability) in most of the input parameters. This is represented by stressors (multiplication factors whose value varies in probabilistic manner as a function of non-engineering parameters like operating condition, culture, weather, and many others). This makes it difficult to accurately model the engineering processes for future predictions using only the underlying physics.
We use a hybrid approach to model engineering processes in the real world. We use the physics/mechanics at the core to represent the system in the ideal world, and then add a layer around it to capture the variability coming from the real world. This layer uses probabilistic formulation using various statistical methods.
The traditional approach of using pure statistics to model the behavior of engineering products and processes may be useful only till the exploration stage. We can understand “what” is happening using pure statistics as we correlate different input variables to the selected output KPIs. As we take the next step to understand “why” this happening is, and “how” to change the system to extract the intended business benefit, we must consider the core science (physics/mechanics) in addition to the statistical tools.
Consider a problem of unintended tire wear that a tire manufacturer is aiming to solve using data analytics. It collects the data related to tire wear, vehicle usage, driver demographics, driver behavior, weather conditions, tire material etc. The statistical analysis will help to correlate various parameters and comment on what is exactly happening. The sample conclusions that could be drawn are of type: “the regions where the average temperature is above 32 deg C, the tire wear is 20% higher than the average”. Or “The tire wear amongst the vehicles driven by young drivers lesser than 25 years of age is much higher than the average population” and so on. As you see, there is a lot of valuable information which is coming out of this analysis, but it is not sufficient for the organization to take concrete steps to change their design or manufacturing process in order to reduce the tire wear. This last step is the real value generation step in which the tire company will get benefitted by lowering their warranty related costs or improving the sales because of better performance of their products over competition.
In the hybrid approach to this problem, we model the tire behavior (physics/ mechanics) under various driving forces to predict tire wear per kilometer of road travelled. This model will work under ideal world scenario. The statistical analysis will complement the core model to quantify the value of stressors to make it suitable for real-world conditions. This hybrid engineering analytics model processes the capacity of answering “what-if” scenarios by changing the tire design, tire material or even the tire manufacturing process. These iterative studies will help the tire company to solve the actual problem of tire wear.
We advocate and follow such hybrid approach in all our engineering analytics projects. This has yielded us great results for our customers. We have a multi-skilled team of electronics engineers, functional experts, computational engineers, data scientists, and big-data experts working synergistically to solve customer problems and unleashing the true value of Engineering Analytics.
This article was from CapGemini: Capping IT Off and was legally licensed through the NewsCred publisher network.