Context:
When I started at Dyson, my team was introduced to the new product concept that they would be working on for the next 16 months. The business was hoping the product would be popular among Asian consumers, whose key requirement is ease of use. So, it included a feature that broke down its main function into 3 sequential phases and used a sensor to automatically switch between phases 2 and 3.
Concept:
I wanted to automate every phase transition by replacing the existing sensor with a new sensing technology that could detect specific events that occurred when a phase transition is required.
These events were too complex for a single sensor to detect without frequent false positives. However, if multiple sensors were positioned around the product, the same events would cause unique patterns in their time-series data. So, my system used machine learning to spot these patterns to avoid false positives.
Development:
There was no prior art for my concept, so I created a rigid toolset to iteratively develop the concept. This toolset was later used by other teams to integrate my technology into new products. The process can be summarised as:
- Design and manufacture a prototype.
- Using it to collect real world data to train machine learning models.
- Deploying the models on the prototype and test performance.
- Analyse the results to understand how it could be improved.
After roughly 20 prototypes the technology was demonstrated as feasible, so I worked with my New Product Development colleagues to integrate my PCBs and fPCBs into the product and create documentation detailing any risks that will need to be addressed before it’s launch.
