All models are wrong ??

It was on a fine Tuesday morning in the year 2020. Well, before you close this article after reading my first sentence, let me confess it was an overstatement. We have all been through a range of emotions in the year 2020, and I'm sure if given a chance, we would all want to wipe it off from our memories. So here I was, seven months into the lockdown, keeping myself occupied with research work and finding my way through the grueling summer of 2020, when the commencement of the Autumn semester provided me a glimmer of hope. I was looking forward to enrolling for a class so that I can get a chance not only to learn something new but also to take my mind off the regular work. I identified a course on dynamic modeling of powertrain systems that was being offered by a reputed professor at the university.

It was the first lecture of the course, and I was pretty excited about it. I had cleaned my desk, prepared a hot cup of coffee for myself, and logged into the online lecture. In the beginning, it was a bit strange but slowly, we got used to the new normal of online teaching. After the initial introductions, the professor jumped into the lecture with a starting slide, having the statement “ALL MODELS ARE WRONG”, written in bold. This statement certainly surprised me and deflated my enthusiasm at the beginning but encouraged me to take a step back and look at the bigger picture.

Fig. 1. Sample results from the Dynamic Load distribution model of a planetary gearset (From my doctoral research)

If you read journal papers, I am sure you would be aware of how researchers constructively find loopholes in the available literature and continually try to improve on them. I am not complaining, in fact I consider it to be a cool banter between talented minds, which will ultimately fuel the advancement of the field. But where do we cut the line? Every model has some sort of assumption associated with it that makes them deviate from physical reality. One could develop a detailed physics-based formulation to explain the intricate details of a physical phenomenon but lose the argument due to assumptions made about the boundary conditions and failing to appropriately capture the external system influences (eg. Fig.1). While on the other end of the spectrum, one could take a system-level approach by ignoring the nitty-gritty details and model each subsystem as a black box governed by simplified empirical relations (Fig. 2). Both these approaches have their set of drawbacks but can be appropriately calibrated to match the physical reality.

Fig. 2. System-level approach to model a Hybrid Electric Vehicle (HEV) Power Train (Courtesy: Mathworks)

Ultimately when it comes to applied research, modeling is done to solve real-life problems, so rather than debating which approach is superior, the right question to ask is, which approach to use when. In conclusion, I would say all models are wrong, but some might be useful based on how we utilize them.