I spoke about the following topics at this Microsoft Conference


Machine Learning for Automotive/ CAE Applications


There are ML applications for CAE that (i) either accelerate by orders of magnitude time to solution compared to conventional numerical simulations and with similar accuracy, or (ii) support post-processing of large amount of data, such as from crash analysis to compare many simulation results at once, and determine what design causes which system response.

To accelerate time-to-solution, Deep Learning can be employed by training a network on data computed by HPC applications, and then infer the solution from input data. Or replace parts of the calculation with a deep network, either with supervised or unsupervised training. This enables a faster what-if analysis for product design optimization.

Data analytics to extract insights from large output data sets may become a  daunting task for humans. ML is useful to better understand the impact of design parameters on automotive parts response to crash. 


Predictive Analytics/Predictive Maintenance (PdM)

This is a prime-time ML application, and yet data availability and quality is critical. Users are advised to start with data telemetry, IoT data and maintenance logs collection, then do condition monitoring and finally build a data-model for failures classification, impending failures prediction, importance parameters calculation, root cause analysis, etc.

PdM requires domain experts, data scientists and ML experts. Especially the feature engineering step is essential to determine good failure discriminating characteristics.

For the Machine Learning model step, there are ready-to-use platforms like Siemens Mindsphere, or Software AG that support partner value add applications, such as SAS.com. Or one can build a ML model based on an open source sample that can be modified to meet specific requirements.

Microsoft also provides Automated Machine Learning for efficient hyperparameters optimization and algorithm selection.

There are success stories such as Audi and SEW that you can view for inspiration and guidance.

Data science and ML modeling is however just one step in the journey: think beyond the proof of concept: will you train your model in the cloud or on premises? Where will you do inference, and possibly using FPGAs? How about data model maintenance? What users interfaces, users training, security, availability and governance should be put in place: remember that performance deteriorates over time and hence model refresh frequency is a critical decision. 

Chatbots in the Automotive Industry


In the automotive market, customer experience is the new competitive battlefield and a primary differentiator. For this purpose, chatbots are a way to reach customers 24/7. Chatbots that implement AI and ML support deep analysis of the voice-of-the-customer, deliver the right message to the right customer at the right time, and systematically transport the desired marketing message to the customer.

But what is an “intelligent” chatbot? Think of the user interface as the tip of the iceberg: what is under water is Natural Language Processing (NLP), Understanding (NLU), and Generation (NLG). The bot also learns from data including customers conversations. And it talks to legacy IT systems, like CRM. 


This enables on the fly context relevant product information and optimized leads assessement. 

KIA Motors uses a chatbot to improve conversions, such as number of scheduled test drives, of credit applications, and ultimately of sales leads.



How would you implement a chatbot in your business? This is a make or buy decision. You could use IBM Watson and develop the chatbot by yourself. Or approach one of the many startups such as botwiser.com or pigro.ai. Whichever way, you will need a plan to maintain the bots, to blend human intervention for un-answered questions, to have a data quality assurance in place, to ensure secure interoperability across platforms,  and to integrate across various channels for connecting with existing analytics.