Background
The Digital Revolution has fundamentally transformed the role that predicting product performance plays in the business model of companies. Customers are increasingly buying directly the outcomes and experiences that result from the product use rather than the products themselves.
We see evidence of this digital revolution all around us. For example, in the Automotive sector, customers are increasingly moving from buying cars to buying mobility. Think for instance, ride-sharing services like Uber, where customers are buying directly moving from Point A to Point B rather than the vehicle that transport them. This has resulted in dramatic drops in automobile ownership, particularly in cities and the young. The result has been that the Automotive industry is making massive investments in Software, Autonomy, Electrification, and Mobility business models. Another example of this transformation is evident in field industries in sectors such as agriculture and mining. In mining, mine operators are demanding up-time and metric ton per hour performance guarantee from OEM equipment providers. These OEMs are scrambling to respond with aftermarket servicing, digitally connected fleets and autonomous operations promises. Even in healthcare, a sector that traditionally has been very conservative, the transformation to outcomes-based frameworks is taking a hold. For instance, the idea of personalized medicine and AI powered robotic surgery is fundamentally transforming clinical outcomes. In this framework, clinical outcomes are more of the result of the healthcare team and associated technology than the individual skill of a medical practitioner. The bottom line is that in sector after sector, customers are increasingly buying directly outcomes and experiences rather than the products that produce them.
Customers are increasingly buying directly the outcomes and experiences that result from product use rather than the products themselves.
This fundamental shift in customer attitudes now requires for companies to not only design products, but to also accurately predict the outcomes these products produce based on actual customer usage and moreover tune, in real-time, the performance of these products to desired customer expectations. Traditionally, engineering simulation software (the predictive tool used by engineers) has been employed to validate product performance and to provide insights that aid in the product’s design. These tools have been typically run by expert analysts with years of experience and in-depth knowledge of their respective domain of expertise. However, the predictions required in this digitally enabled world need to be fast, require very little overhead or infrastructure (i.e. can run locally without massive communication or computational hardware), must be simple (i.e. cannot require a PhD to run them), and need to be adaptive to changing customer usage and requirements. Engineering Simulation as it stands today is simply not up to this task.
Predictions required in this digitally enabled world need to be fast, require very little overhead or infrastructure (i.e. can run locally without massive communication or computational hardware), must be simple (i.e. cannot require a PhD to run them), and need to be adaptive to changing customer usage and requirements.
Physics Informed Machine Learning (PIML™)
Our patent-pending Physics Informed Machine Learning (PIML™) technology converts complex and time-consuming engineering simulation workflows into fast running instantaneous solvers. Our detailed PIML™ algorithm is proprietary. However, in general, this algorithm uses physics-based simulations to capture what we call the critical path response field creating multistage Reduced Order Models (ROMs). The algorithm then utilizes Data-Driven Machine Learning techniques to calibrate the prediction. The result is that PIML™ in contrast to purely Data-Driven Machine Learning techniques, is much more accurate, requires significantly less data, gives users insight as to how to improve the solver’s predictive capability, is more robust to changes in the environment and can be integrated into Bayesian information fusion frameworks.
PIML™ in contrast to purely Data-Driven Machine Learning techniques, is much more accurate, requires significantly less data, gives users insight as to how to improve the solver’s predictive capability, is more robust to changes in the environment and can be integrated into Bayesian information fusion frameworks.
Figure 1 illustrates key benefits of PIML™ technology.
One of my colleagues (Mr. Sreekanth Gondipalle) wrote an interesting article detailing a case study (linked below) showing the transnational performance improvement of Physics Informed vs. Data Driven Machine Learning methods.
PIML™ is particularly powerful when the input/output relationship of parameters in the model are complex and when data needed to train this model is scarce. Figure 2 illustrates the applicability of PIML™ vs. Data-Driven Machine Learning techniques. PIML™ surrogate solvers provide a unified computational engine that powers and transforms both Simulation Powered Design and Predictive Digital Twins. PIML™ solvers can be used as the back-end computational engines of Simulation Apps. They can also be embedded as components in System Performance Simulation tools or integrated into PLM/SLM/ERP/IoT platforms. Using PIML™ as a solver as opposed to automated engineering simulation workflows, enables designers to get real-time feedback to design changes without the massive software licensing and hardware infrastructure required by most engineering simulation software. Figure 3 illustrates how PIML™ is employed for both Simulation Powered Design and Predictive Digital Twins.
PIML™ surrogate solvers provide a unified computational engine that powers and transforms both Simulation Powered Design and Predictive Digital Twins.
Digital Monetization Strategy
When devising a digital monetization strategy many people focus on global technology trends and design the monetization strategy based on the future end state. However, there are low hanging fruit digital monetization opportunities that can utilize your existing business models and distribution infrastructure. Therefore, a good digital monetization strategy should look for opportunities to monetize your digital infrastructure in the short-term utilizing your existing business model and distribution channels, as well as, in the longer-term period, where fundamental changes may have occurred to your business model and distribution infrastructure.
A good digital monetization strategy should look for opportunities to monetize your digital infrastructure in the short-term utilizing your existing business model and distribution channels, as well as, in the longer-term period, where fundamental changes may have occurred to your business model and distribution infrastructure.
Digital Monetization By Leveraging Traditional Business Models
There are many ways that PIML™ Monetizes Digital through your existing business model and distribution infrastructure. Below are a few use cases:
1. Simulation Democratization: PIML™ powered surrogate solvers can be integrated into your existing simulation powered design democratization initiatives. Traditional automation-only strategies for simulation powered design democratization initiatives have significant limitations. These strategies still require for designers to run the hardware intensive and time-consuming engineering simulation with their associated costs and inefficiencies in order to explore the design space. PIML powered surrogate solvers on the other are fast, adaptive, and do not required all the infrastructure of the automated engineering simulation workflows.
2. Sales Accelerators: PIML™ can also be used to accelerate product sales using your company’s existing sales and distribution lines. PIML™ enables companies to create complex configurators and demonstrators that can be web-enabled or enhanced through virtual reality (VR) showcasing to customers how your products will perform under real-world complex scenarios. This enables customer to better choose the product configuration that fits their actual usage needs. Most point of sale configurators today rely either on simple physic or gamification techniques that cannot accurately predict complex realistic usage scenarios.
3. Warranty Claim Occurrence Predictor: PIML™ could also help reduce your company’s warranty claims. We recently demonstrated to a customer that we could feed field data to our PIML™ solver and accurately predict where a warranty failure would occur. Companies spend big money doing all kinds of statistical analysis on warranty data trying to predict and make bets on where they are likely to have a claim. If you can predict where these will occur more accurately than these data-driven methods, then companies can make better bets on their warranty contracts.
4. Model Validation & Good/Bad Data Sorter: PIML™ can be used as classification tool to distinguish “good” data vs. “bad” data say in a test environment. This can significantly reduce your prototype test costs and make your test more effective. PIML™ can also help you determine what to measure and what type of sensors are needed to digitize your product. We recently showed a customer that with only two sensors they were already using, we could predict the performance of the machine for fatigue damage. This capability we call virtual sensor powered by PIML™.
5. Model Based Controls: PIML™ speeds models to make them usable in model-based control architectures that require models to run faster than real time in application such as system performance simulation, autonomous systems, climate control, to name a few. These model-based control architectures can then eventually be incorporated into Digital Twins
6. PIML™ together with our consulting services can be a catalyst that accelerates your digital transformation. Companies and especially large companies have silos, organizational politics, and competing digitalization visions. Executive Management spends an inordinate amount of time and resources trying to align and execute the vision and strategies of the company. PIML™ can help align organizations, bridge the gap between silos and help create shared execution commitment. Because PIML can be used as the solver system for both your development groups simulation powered design democratization initiative and operationally by your digital twin digitalization efforts, companies now have a “common brain” that can speak the language of their cost centers, as well as, the language of their revenue centers.
Digital Monetization by Leveraging New Business Models
The ultimate expression of a digitally enabled outcome-based economy is Product as a Service or PaaS. In this framework there are no products, just the service outcome they provide. In such a model, products essentially become service robots that adapt to desired experience and outcome needs in real-time. This will be the age of Digital Twins and their associated Internet of Things (IoT) infrastructures. Here, PIML™ powered surrogate solvers can be embedded into onboard processors creating Predictive Digital Twins that can provide in real-time the response of key performance parameters to input operating conditions. These Predictive Digital Twins would incorporate not only simulation data, but also test and field data using Bayesian Information Fusion techniques.
PIML™ powered surrogate solvers can be embedded into onboard processors creating Predictive Digital Twins that can provide in real-time the response of key performance parameters to input operating conditions.
There are multiple applications for these Predictive Digital Twins, and some are illustrated in Figure 4.
How to get started?
You can find out more about PIML™ by downloading our overview presentation from our website. You can also contact us to learn how we can help you monetize digital and ensure your engineering meet dollars!