Ariston Thermo Group (Nord Clan: ML Sense)

2024: Machine Vision for Home Appliances Quality Control: Implementation at Ariston Plant

The Ariston electric water heater plant in Vsevolozhsk produces more than 600 thousand devices per year. Sometimes they face a problem in production: experts do not notice small defects on the tubes of water heaters. To prevent the production of defective products, it was decided to use machine vision technology.

The challenge from Ariston is to automatically detect defects in the hot water intake tubes of a water heater using machine vision technology.

Peculiarities of the customer's requirements to the defect control system

1. The conveyor should automatically stop as soon as the system detects a defect: no insertion of water heaters in the tube or burrs. In this case, sound and light alarms should be triggered.
2. Depending on the model, the distance between the tubes and the height of the water heater on the conveyor may vary. And the water heaters can be radially displaced on the conveyor.
3. Solve the problem turnkey: design mounting masts and install the necessary equipment: machine vision cameras with backlighting, server equipment with a monitor, light and sound columns to notify the operator of defects.

Stages of automation of quality control of household appliances

1. Production test video to identify control points for installation/equipment. It was important to assess the factors that affect the operation of the ML Sense system - the level of lighting, the presence of vibrations, the radial displacement of water heaters on the conveyor, the difference in height between water heaters and the intended chamber.

2. Selection of chambers that will cope with the search for defects on the tubes of water heaters. Selection of lighting devices with high-intensity LEDs to obtain clear high-contrast images for neural network recognition.

3. Testing of the quality control system of ML Sense water heaters tubes in its own laboratory.

4. Simulating a virtual 3D scene based on measurements from a production line. Calculation of optimal distance for installation of chambers, lighting fixture and preparation of mast drawings for equipment attachment. Fabrication of the structure.

𝓲

Nord Clan

5. Training of the neural network to recognize typical tube defects. To do this, we collected a datacet from photographs, where each type of defect is marked and classified: there is an insert on this tube, there is a burr on this one.

To quickly notify employees about defects, a boxed notification system was introduced. Three warning classes were assigned to defects: red - no insertion on the tube, yellow - a burr on the tube, green - a tank without defects. As soon as the system "sees" the defect, an audible signal is triggered and the conveyor stops. The controller can remove the tank from the conveyor, which does not have an insert on the tube, or cut off the burr.

The equipment was installed at the Ariston plant in Vsevolozhsk. They installed masts, fixed cameras, installed software at the control post, trained personnel to work with the ML Sense system. Completed commissioning.

𝓲

Nord Clan

Already in production, during the work, a new type of bushing was revealed - metal. There was no bushing in the customer's terms of reference, so we initially trained the system only on a typical enameled bushing. But since we are always for the decision to work and benefit, we have finalized the system and further trained the neural network.

𝓲

Nord Clan

Implementation result

The system in 100% of cases sees a defect, notifies the operator about what type of scrap is detected, by sound and light signal, stops the conveyor.

The conveyor line operator is spared routine labor and works faster. The employee can only remove the defective product from the conveyor, send it for revision, or fix the defect manually.

As a result, the economic effect is increased - you no longer need to risk the ruble for complaints about defective products and the company's reputation due to the human factor.