December 2020 • Issue 2
NNAISENSE delivers advanced neural network solutions for industrial clients that improve how their processes and products work.
AI in Industrial Applications
Business benefits in real deployments, customer success stories
Interview with our customer and strategic investor TRUMPF GmbH
In short, we found only two companies having the potential to solve our machine case and only NNAISENSE has proven industrial evidence to understand how to work out a solution with the given methodology.
Dr. Dieter Kraft
Managing Director, TRUMPF Venture GmbH
NNN: Why is TRUMPF so interested in AI solutions?
DIETER KRAFT: AI is part of the digital transformation at TRUMPF and our innovation promise to our customers. We are already using AI in a multiplicity of use cases.
What are the use cases that you foresee, and what business benefits do you expect?
The use cases are broad based. Let me start with manufacturing machines for sheet metal production. In this business area, we are using a data lake concept with data collected from machines being used to enable them to learn as a network node on the edge. A potential next step could be to finalize the fully automated machine concept together with NNAISENSE by solving the missing link of sorting out the different parts into the right storage box at the end of the production process. This task sounds easy, but actually relies on the vast knowledge of people handling such parts today to cope with the possible variation in part geometry, material thickness and other parameters. The business benefit, once this task is solved within process time and with excellent yield, is the introduction of this AI solution into our Smart Factory setting. Another use case is related to services and predictive maintenance which involves online adaptation of machine parameters in response to intrinsic sensors and material state
- Interpretation of written service protocols
- Automated recognition and order generation including storage check for requested spare parts
- Scheduling an optimized customer route for the service technician and report generation in the connected CRM system.
The business benefit here is clearly customer satisfaction due to minimized idle times and optimized service up to customer self-services in a still restricted application area. A bit more in the future, but on the road-map, is distributed learning to collectively benefit from the machines in the field and redistribute such learnings while maintaining data privacy.
Why did you choose NNAISENSE as your AI solution partner?
The answer is on the one hand simple, but on the other hand not that obvious. Let me first elaborate on the simple part of the answer. In short, we found only two companies that have the potential to solve our machine case, and only NNAISENSE has the proven industrial experience to understand how to work out a solution with the given methodology. Even though the claim of “we can solve it all” is pervasive in the AI space, there are only very few companies that offer the relevant technology and understanding in their team to play out such unique selling positions. Given the competitive landscape of offerings, this is the part you need to analyze deeply in order to understand the technical approach and also the team. This means committing the necessary resources to diligently evaluate the performance of the team both directly, and by talking to people who have already succeeded with the NNAISENSE approach.
Why did you become a strategic investor?
TRUMPF: As the corporate venture capital arm of the TRUMPF Group we are interested both in bringing strategic benefit to the parent company, and also getting risk-adequate returns on our investment. As a strategic investor, we intend to provide more value than just capital. Therefore, we will be very pleased if we can add the “Proof of Concept” from our industry to complement those of the industrial partners who have already experienced the performance of the NNAISENSE team. When we see a good fit with the team, investor syndicate, technology and financial return potential, investment is a good vehicle through which to gain deeper insights into a company, broaden our own knowledge and support them in other areas if we can.
In general, how do you see the state of the German high-tech industry and AI adoption, in particular compared with the US and China?
I think the answer is: it depends. The state of the German-based AI industry splits into different application areas. The area of photonics-based light field interpretation and pattern recognition is well established and was one of the first AI application areas. Internationally, the market is already fragmented with many deep learning offerings from all around the globe. Application areas with special domain knowledge, like those requiring high-end technology or more advanced AI approaches, such as reinforcement learning or combinations of multiple pathwaysas offered by NNAISENSE (and at least one other competitor), are still at an early stage of deployment.The US and China are usually thinking bigger than German or EU companies in terms of scale and level of adoption. However, I personally do not think every good technical software solution must be based on AI. Only certain dedicated tasks need to be. Meticulously analyzing a challenge and working out the respective problem space with an adequate solution is not an easy task, but might be the sustainable way to build a company and its long-term business success.
Updates from the Research Team
Artificial General Intelligence Part 1: On Long-term Safety in General Intelligence.
This is the first of a three-part contribution from our AGI team who developed the AIOS-3 platform.
Event presentation, IJCB2020
Encouraging benchmarking performance with a Clustered Dynamic Graph CNN for Biometric 3D Hand Shape Recognition
“With the current difficult hygiene situation, we believe that this new approach mitigates the problems of the contact-based biometric devices.” – Jan Svoboda
Read paper
Learning to detect objects with a 1-megapixel event camera, NeurIPS 2020
We’re delighted to have been involved as co-authors with the PROPHESEE team in their brilliant work, and look forward to making further advances in this area together.
Read paper
How to Learn a Useful Critic? Model-based Action-Gradient-Estimator Policy Optimization, NeurIPS 2020
We developed a new algorithm for actor-critic reinforcement learning, which explicitly trains the critic for value gradients w.r.t to actions using temporal differences.
“This is a fundamentally new way of using learned dynamics models in reinforcement learning for continuous control.” – Wojciech Jaśkowski, Senior Researcher and RL specialist at NNAISENSE
Read paper
AI Industry News
IoT, Industry 4.0, manufacturing, industrial processes
PwC report: What’s the real business value of AI and how can you capitalize?
The three areas with the biggest AI potential in manufacturing are:
• Enhanced monitoring and auto-correction of manufacturing processes
• Supply chain and production optimisation
• On-demand production
