Predictive engineering analytics is about more than just big data
14 March 2018
Big data has been hailed as ‘the new oil’; a frontier that you can mine for insights and forecasts. We’ve moved from looking at data that tells us what happened, to why it happened, to what might happen next.
Predictive engineering analytics combines physics-based simulations with data mining, statistical modelling and machine learning techniques, using patterns in the data to build models of how the systems you gathered the data from work. With those models, you can discover what the data you have tells you about the data you don’t yet have.
IoT and sensors are already transforming products, and mining the stream of information from products will be critical for maintaining products and designing their replacements – but that's not the only part of product development where predictive engineering analytics matters. And if you’re looking at predictive engineering analytics as relying only on big data, you’re missing some of the key opportunities.
For many industries, the products they create are no longer purely mechanical; they’re complex devices combining mechanical and electrical controls, and functioning in ever more complex environments. That means engineering different systems, and the ways they interface with each other, and with the outside world. At one level you’re coping with electromechanical controls, at another you’re creating a design that covers the cooling requirements for the electronics. And in future, you have to model that as part of a larger systems; for instance, systems inside a vehicle will begin to talk to other vehicles and to traffic systems on the roads they travel on.
One consequence of this increasing complexity is that testing during engineering has been routinely supplemented by and, in some cases, even replaced by, simulations that cover multiple systems, and take into account the many different types of physics you need to model all those systems. That’s valuable during design and in acceptance testing too. Either the physical product design or the demands of the location of the finished product may make it impossible to gather readings from a physical sensor to verify final performance. That’s when a virtual, simulated sensor can augment the information from the physical device and enhance the usefulness of the test.
On the other hand, demands for strength, fuel efficiency or simply more efficient manufacturing may mean adopting new types of materials and new production techniques, instead of relying on well-known ones. Companies who have decades of experience with traditional materials like steel and aluminium have to learn to work with new materials, often using additive manufacturing and combined additive and subtractive manufacturing. That means going back and doing physical tests and correlating those tests to simulations to understand things as basic as how materials behave at a range of temperatures and what impact that has on the system design.
To address all these demands, companies will need to integrate their testing methods and their simulation methods, and they’ll need to adopt both more simulation – and much more data management, so they can accelerate the speed at which they perform their engineering work. This goes far beyond tracking requirements, CAD data and test results; engineering data management systems need to store all the engineering work, including simulation and verification, and integrating test, sensor and performance data.
That becomes even more important as the trends of mass customisation and personalisation increase, making it impossible to test all the different versions of a product in all potential environments. Instead, you need to be able to leverage simulations to get broader coverage of all variants and usage scenarios. That allows you to go back at any point and prove that you verified a component with all the relevant systems, to dig into the full range of data to see what might have led to a failure and to use predictive analytics to forecast how products will perform.
The mathematical approach of big data analysis is certainly useful. But applying predictive analytics to the engineering space needs an approach that combines test data and physics-based simulation data in a common database environment (and not just a single type of physics but multiple types). That allows engineers to take very large test and simulation data and use them to create key performance metrics.
Predictive engineering analytics also covers exploring the design space efficiently by running multiple simulations with different parameters and analysing the resulting data intelligently, so you can understand key parameters and how they interact. That enables you to optimise a design to achieve robust performance that’s not sensitive to changes in the environment.
Predictive analytics may even move into products, as control systems shift from detecting to forecasting conditions. Today, anti-lock brakes use sensors to detect when the car is beginning to skid. In the future, sophisticated control systems could use on-board cameras to detect that a vehicle driving in the rain is approaching a curve too fast for the road conditions, predicting that it will skid and controlling the vehicle before it does, to handle the curve safely.
The size and scope of the data sets available enables advanced analytics, but this size also has consequences: engineers will be drowning in information unless they take steps to make big data manageable. The number of sensors in products today is only going to increase. Sensor readings from an aircraft engine or measurements taken from a car on a test track already create huge data sets – too large to use the raw data in simulations, because they’d take too long to process. The scale of the data is going to require intelligent analytics to condense raw streams of readings into data that can usefully be fed into a simulation.
Predictive analytics is already useful in engineering today. You can use it with your simulation portfolio to investigate different architectures early in product development, to help you understand which type of architecture is best suited to meet customers’ needs. You can create 3D simulations and integrate them with those early architecture models. Then you can bring in test data and see how it correlates with the simulations to improve your models and increase the fidelity of the systems. As software tools develop, you’ll be able to simulate new types of physics to improve the accuracy of your models, simulate large systems and more complex models, and to take advantage of more analytics tools.
You can also benefit from integrating analytics into other tools. Having the information from those simulations in a central information system enables teams beyond the engineering organisation to understand architectural decisions and use that information in their own processes.
Combining data from devices with warranty information and data about customer satisfaction in a common data store allows you to use big data analytics to understand the significance of different factors in the design of the device that ultimately have an impact on business success. If you do that in a PLM environment, you can integrate many different types of data in a more connected way; as the models, simulations and test results are updated and new customer information comes in, the analytics can take account of those to give you an up to date view of the situation.
Many companies take advantage of one or more of these opportunities today. But if you want to really make the most of predictive engineering analytics, you need to look at this more holistically. If you bring all your information sources together in a more integrated system where you can apply new and emerging technologies as well as your existing tools, you’ll get more value out of them – and you can more confidently engineer products that perform well in complex environments.
By Jan Larsson, Senior Marketing Director EMEA at Siemens PLM Software and Ravi Shankar, Director, Global Simulation Product Marketing at Siemens PLM Software
Contact Details and Archive...