Testing the output quality of electric window controls for cars
In the past operators tested the quality of the window regulators via listening. This meant the testing approach was dependent on the human factor and listening by ear. The manufacturer decided to replace this manual and demanding method by automatic system delivered by Neuron Soundware.
The challenge: replacing the human operator who makes complex decisions 100s of times per day on the production line
Top market leader in the automotive industry produces Mechanical Control Cables, Window Regulators, Door Modules and Power Closure products. With facilities in many countries worldwide, the producer is represented in all major automotive markets globally.
Annually producing over 10 million automotive cable applications and window regulators, the producer also remains one of the top market leaders in cable applications for recreational vehicles such as boats, ATV, personal watercrafts, motorcycles, and golf carts.
The quality of the parts produced is important for this leading manufacturer.
As part of the production process of window regulators, in the past operators tested the quality of the window regulators via listening.
This meant the testing approach was dependent on the human factor (listening to each window regulator) at the end of the production line, a highly subjective procedure which resulted in a high number of claims from customers.
For this reason, in 2018 the manufacturer decided to evaluate the suitability of Neuron Soundware (NSW) sound detection technology in order to increase the accuracy of the quality control testing on one of their car window mechanical system assembly lines. It was agreed that the project will take place in the facility located in the Czech Republic.
The goal of the project was to develop a solution which will be part of the quality control end of line testing of the automotive mechanical unit developed by the manufacturer.
Before the start of the project, NSW received data from the manufacturer in order to assess the feasibility of the project on a small dataset. The project proved to be viable, so NSW proceeded with additional data acquisition, training of the machine learning classification algorithm used for the task and deploying the service to the existing platform and quality control system.
The solution: Acoustic quality control on production lines using artificial intelligence
Today, approximately 1600 products are being tested with the NSW AI solution every day. The solution consists of IoT devices that are processing data, the Arduino for providing information about the running test, and the ML model that evaluates noise levels throughout the whole frequency range and provides objective data-based results in a matter of seconds. As the process is AI-based, it has the ability to adapt to changing conditions and learns over time.
The products are tested on a test-bench equipped with 3rd party sensors, where the process of moving and retracting the window takes place. The manufacturer tests several parameters, sound being one of them.
For sound, the ML model from NSW evaluates the entire curve at all frequencies and determines whether the set limits have been exceeded. If this happens, a red light on the screen informs the operator that he/she should send the product for another inspection.
The data is stored in the cloud. The employees responsible have access to the nGuard portal where they see data about all tested products and can identify the product according to its Barcode.
This is important in the event of a complaint, as it is possible to download the sound sample and its evaluation, thus providing full traceability for each individual product. NSW also provides the analysis of the soundtrack against the set threshold involved in these circumstances.
What the inside of the NSW nGuard app looks like
In addition to the screen directly in the production facility, the customer also has the nGuard application environment at his disposal in which he can see the tested products over time. The image below shows that 77 products were tested between 19:00 and 20:30. Of these, 74 were fine and 3 products were evaluated as defective requiring additional manual inspection.
A mathematical model loaded on a NSW IoT terminal located in the customer’s production facility evaluates for every recorded sound sample its pitch for each of the 20 required frequencies. (For each frequency, the volume must be within a certain standard according to the given maximum permissible volume level.)
Conclusion: Neuron Soundware solution
Before the NSW technology was deployed, human operators made complex decisions on the production line up to 700 times a day. The NSW solution enables the transition from subjective manual testing by human listening to objective standardised automated testing.
The main advantages of the deployed system are:
- reducing testing costs,
- significant increase in workplace health; minimising workers’ exposure to hard-to-hear sounds
- greater accuracy, evidence and objectivity in product quality evaluation
- historical records of testing, the possibility to compare results retrospectively with each other
- quality certificate for each finished product based on measurement data
- reducing the cost of complaints from the customer to 0%