| The Renault LHA Sandouville factory, located
in the Haute Normandie region in the port and industrial estate of Le
Havre , is the main production plant for Laguna, Vel Satis and Espace
IV. The site includes four manufacturing departments: pressing; sheet
metal; paintwork; and assembly; along with eight departments and sections
carrying out support functions. It has a daily production capacity of
1,900 vehicles. It was the Renault Group’s first bodywork assembly
plant to receive ISO 14001 certification in December 1998.
A vision application was set up in the pressing workshop that produces sheet metal parts that make up the bodyshell of the vehicles. It mainly comprises cutting lines, assembly and sheet welding points as well as sheet presses which make up the various bodywork components. The pressing workshop supplies the sheet metal workshop which carries out the assembly of the various parts of the bodyshell. This then passes on to the paint shop before final assembly. The vision system installed is designed for checking the right and left sides of the bodyshell of the Laguna cars. The same installation is therefore able to check four different parts.
Each side of the bodyshell is made up of two parts which have been assembled by laser welding. It is around the weld that defects can appear and need to be detected. While the various pressing operations – which give the part its shape – are being carried out, clear or partial breaks can be produced. As the parts are stored manually at the end of the line, those parts showing such faults are easy to spot and remove.
It is not so easy to spot when the faults are small holes, some of which can measure less than 0.3 mm which form along the bead of the weld. If the presence of these holes is not detected at the end of the line, and if the part is not removed, breaks can be produced during shaping, causing numerous problems which can have serious effects on the productivity of the assembly line. Sometimes minute holes are enough to give the bodywork an unsightly appearance after the paintwork has been done.
If these defects are not detected before the vehicle is assembled, the consequences can be financially very expensive as it is very difficult to repair – and sometimes there is no other choice but to send the chassis to the scrapyard.
Checking the welding was previously done by several operators who had to struggle to handle parts of a considerable size and weight: the sides of the bodyshell measure around 3.4 metres in length, 1.6 metres wide and weigh nearly 30 kg. To carry out the operation, they used to place a light source on one side of the part and check from the other side that the light did not pass through, showing breaks or holes in the part. This form of control, even when it was done in the optimum way, cannot highlight holes of a very small diameter.
Numerous defective parts passed through these checks, so a solution had to be found which was capable of putting a stop to this situation while carrying out a continual and reliable check of the lower bodyshells.
At Renault, Patrice Dumont is the automated systems manager. He is part of the DIVD (the decentralised vehicle engineering department) pressing section, attached to the Renault Technocentre.
Some time before, Dumont had the opportunity to attend a seminar given by the machine vision company Cognex. He realised that vision systems, which are used widely in industry were a better alternative to other control procedures, and that they could be the solution to the problem. They reduce the difficulty of certain tasks that are normally carried out by operators. Not being affected by tiredness like the human eye, they continually work guaranteeing quality and improving productivity.
He decided to evaluate a solution based on industrial vision systems and an integrator of the Renault Sandouville factory was called in. It was necessary to be sure that this type of check could be suitably carried out with a vision system and a prototype could be quickly produced for a first test. Several vision systems manufacturers were called upon. Cognex was the most reactive, and it was the In-Sight vision sensors range which were chosen for the operation.
In July 2002 the prototype was tested. The trials proved very conclusive: the tests carried out were 100% correct. The Cognex In-Sight vision sensors detected all the faults. Terms and Conditions were drawn up and a first test bench was ordered. It was specified that the system should be capable of detecting holes with a 0.3 mm diameter. In reality, the system would demonstrate that it could do better than that.
The first test bench was put into service around mid-2003. A second test bench, with smaller dimensions, was ordered later for checking the side frames of several vehicles (Laguna, Vel Satis and Espace) on another line. Each test bench has four Cognex In-Sight vision sensors.
The body shell side test bench
The first test bench is a metal frame around 5 metres long and 2 metres wide, weighing 5 tonnes, with a 5 cm thick platform holding the structures on which the parts to be checked are placed, along with a backlighting system with LED indicators. The test bench was put in place by a bridge at the end of the pressing line each time the manufacture of the Laguna lower bodyshell is started up.
Four Cognex In-Sight 1000 cameras, each placed in protective casing, are positioned at the top, over the platform, on the cross support beam. A control screen is fixed on one side of the test bench, in a box where the control system is also installed. A marking system for defective parts is fitted at the other end.
The parts to be checked – bodyshell sides – measure around 3.4 metres long, 1.6 metres wide, and weigh nearly 30 kg. They are put onto the supports provided for this purpose on the test bench by a robot. There are two areas to be checked. The target areas measure approximately 10 centimetres in length by 5 centimetres wide, with a covering. Two of the cameras are inclined at 45°, the part forming a U-shape over the welded area.
The cameras are able to detect very low light levels coming from the backlighting system and passing through any holes, some of which are only a tenth of a millimetre in diameter.
If the part is declared sound, this is displayed on the control screen. If the part is defective, it is also displayed on the screen, a red light comes on at the end of the line for each fault, and finally, the part is marked by a jet of ink which stops it from being used.
The speed of the control carried out on the line was a determining factor of the project: all the operations – positioning the part, capture and analysis of the image, detection and marking – could not reduce the production rate in any case. At this stage, the Cognex systems proved their worth: the production rate specified in the terms and conditions was 900 parts per hour; the current production rate is 420 parts per hour for the bodyshell sides and 850 parts per hour for the side frames.
The system is connected to a PC fitted with a hard drive for saving the photos of the defective parts detected over one year. This procedure allows the problems encountered to be analysed at a later date; it contributes in ensuring the traceability of parts and for monitoring the various shifts.
Simplicity of operation
The aim of this project was to find a method and a system for automating the checking of welds. The principle of the checks relies on the backlighting of the area to be inspected. Resorting to a camera for seeing if rays of light cross through the welded area was logical. However, in order to “industrialise” the solution, there were still several points to be checked: firstly that the tools for processing the images acquired were efficient enough – precise and reliable – to recognise all the types of faults, including those difficult to detect with the naked eye, and secondly to carry out a 100% check without reducing the speed of production.
“We had to teach the system what faults were to be identified”, says Dumont . “That was done progressively at the same time that we were familiarising ourselves with the system. Once the system had memorised the type of fault, the recognition rate was 99.99%.”
The main concern which had to be overcome concerned the ambient lighting. Because of the existing lighting in the workshop, the layout of the area (glazed surfaces) and the orientation, reflections appearing on the parts occasionally disrupted the operation and showed faults – which were nonexistent. Placing curtains at the end of the line resolved this problem.
The welding checks are now carried out reliably on the whole production (bodyshell sides and side frames). With the fault recognition rate obtained, the defective parts are very quickly identified and removed at the end of the pressing line. “An end to the time when entire bodyshells had to be sent to the scrapyard”, says Dumont . “The savings made at this stage alone fully justify the investment. The fact that the vision control point is shared by four parts also contributes to the increase in its profitability.”
This method of checking welds with industrial vision tools is of interest to other sites and is likely to be brought into general use. A patent has been registered covering the principle used.
|
|
|
