Outcomes, Insights and Best Practices from IIC Testbeds: MQM Testbed
process and maintain a non-intrusive
approach to test the product;
2. An automated objective measure for
quality check shall be used;
3. The result must be at least 30%
better than the status quo.
A LTERNATE S OLUTION
It is typical that when a factory process is not
significantly broken, it is often left intact as
long as the throughput of the products met
the minimum requirement: Changing the
layout or moving the equipment around will
cause direct impact on the production flow
and production schedule. Although the team
found outdated quality control processes
were used to detect defects making it an
obvious target for improvement, the design
team felt a different tactic was needed to
convince the client the change was
necessary and that the new solution would
greatly improve the status quo. The team
was surprised to see the quality check of the
whole process relied on the judgement from
experienced professional examiners to listen
to the noise when the air-conditioner was
turned on at the end of the assembly line.
The decision whether to pass or fail each air-
conditioner was made by the listener’s
subjective decision. Although the method
looked obsolete, the client seemed to be
content with a three-listener rotation team
to perform the task. That made the
proposed change more challenging as the
new solution needed to be easily
understood by the client and the results
would be required to be significantly better.
The only complaint the existing process drew
was, sometimes, the fail/pass (faulty
product being passed) rate was higher than
normal and, as a result, the management
and the logistic team was not happy when
many products were returned after
shipment.
Furthermore, it is necessary to find a quick
solution to meet the timeline of the MQM
Testbed deliverable dates since three
months had gone by. The task seemed
straightforward, replacing the listeners with
a machine, but the devil was in the details.
Although the most discrete characteristic of
the manufacturing lines involves welding,
the new focus of the MQM Testbed was on
improving the quality control through
acoustics analysis and analytics of the end
products. The core of the experimentation
was around how IoT could be applied to this
environment and how analytics for acoustics
could improve the accuracy rate of correctly
detection of defected products.
The design team went back to the lab and
start redesigning the MQM Testbed based
on three requirements: There are three major sections in the IIRA:
sensors and sensory network, analytic
platform and management. The MQM
Testbed uses an analytic engine to perform
the computation and assessment of quality
control to determine the pass/fail of the
product. Data must be collected in the field
to train the artificial intelligence (AI)-based
analytic engine. The process needed to be
fine-tuned to ensure it can be used for the
design of the production line. While
designing the MQM Testbed, picking up the
acoustic signal and the data presentation for
control
and
management
were
straightforward. However, it may have been
problematic to find a useful analytic engine
that can be trained for acoustic noise
detection.
1. The testbed shall minimize the
disruption of the current production Fortunately, there was an AI-based analytic
engine solution, based on technology from
- 24 -
March 2018