Outcomes, Insights and Best Practices from IIC Testbeds: MQM Testbed
Huawei, which enables a system to pick up the field signals and train the system, based on the collected data and previous results. This leads to correct decision-making based upon the preset policies and rules. This analytic engine was originally built for another project. The team was lucky to have it modified for the MQM Testbed. The only criteria communicated to the AI – based analytic engine development team was that the analytic engine must make the right call with the focus on avoiding faulty detection by passing a product that is actually defected. Fortunately, the analytic engine was easily modified to suit this purpose. If not, additional delay may have occurred if the testbed team needs to start over.
AI-BASED ANALYTIC ENGINE
Up to this point, people may question using an AI-based analytic engine for this acoustic detection task as overkill. The process may take place with a dedicated machine and does not need to involve many computations. An AI-based analytic engine, such as the one provided, performs machine learning and deep learning, which requires not only the computing power of an ARM processor 2, but also some other processors like Graphics Processing Unit( GPU) and CPU to support it. This was a conscious decision by the testbed team because it could help yield an additional benefit for the MQM Testbed: with the AI-based analytic engine, the MQM Testbed was not only very useful for the simple acoustic detection task, it could also be used for other purposes in the future.
It is important to note that the training data sets for the AI-based analytic engine may not be easily acquired. The key to machine learning is the training data sets. Project partners provided field recorded data from machines in good working order( passing data) and the machines being returned( failing data) after shipping. Colleting the failure data sets was not an easy task because the quantity of the failed machines is relatively small for the short data collecting period. As a result, the data set size of the failing data is much smaller compared to the size of the data set for passing data. All data sets are fed into the learning process of the machine to train the machine to identify the difference between passed and failed states. The machine eventually established a passing state and failed states through testing, training and fine-tuning. This analytical engine development is very important as the AIanalytic engine is not something available off-the-shelf. For users who plan to modernize an existing plant, the complexity of the project cannot be underestimated.
TESTBED COLLABORATION
One challenge of promoting the MQM Testbed is that colleagues from Haier have been unable to travel to IIC meetings to share their knowledge with members. The testbed spokesperson has communicated MQM Testbed conversations to IIC members with the goal of developing this testbed further within IIC. While other IIC members may have similar analytic engines, there are aspects that are unique to Huawei’ s approach for the Testbed as IIC members are able to duplicate the process and develop
2
An ARM processor is one of a family of CPUs based on the RISC( reduced instruction set computer) architecture developed by Advanced RISC Machines( ARM). TechTarget, How to choose the best hardware for virtualization.
IIC Journal of Innovation- 25-