Accurate data closing the manufacturing loop
29 May 2007
Peter Conlon explores the changing role of AOI in the manufacturing loop.
These days Six Sigma or continuous improvement systems are typically employed and neither of these would be possible without timely access to accurate process data. Speed of decision making will govern how much of a production lines' work in progress (WIP) is within acceptable quality parameters. If a manufacturer only reacts to a quality issue when a product is returned by a customer then the number of products with potential quality issues could be enormous. It's therefore obvious that quality decisions should be made as early in the manufacturing process as possible.
Process data and its accessibility then becomes a function of two requirements of that data;
1) the speed at which the right data (or the currently relevant data) can be presented when a decision has to be made and,
2) the accuracy of the process data used in decision making.
In electronics manufacturing it is typical that a number of measurement and test technologies are employed throughout the manufacturing line. AOI, AXI and ICT are typical examples of which, some or all are found on almost all SMT production lines. AOI and AXI allow the inspection of each and every component on every assembly manufactured. If a mobile phone circuit board is taken as an example then this may mean 2000 devices per panel each with 5 AOI inspection results, with one of these panels being produced every 15 seconds. That¡¦s 40k data point per hour or approximately 840k per day. It should be noted that this is only for the actual inspection itself; there may be images of failed devices and repair results to store as well. It would be easy to drown under a sea of data.
With such a volume of data finding the causes of problems on the production line could be very difficult if not, almost impossible. This is particularly the case when you imagine the speed of production on a modern SMT line. Process engineers need to react quickly to ensure that the line is producing a quality product and not scrap.
AOI machines inspect and measure each component on a SMT assembly. The primary source of information from AOI is its inspection data. However there is a very valuable set of secondary information that being the attributed data, associated with each component. Each component on a SMT assembly has to be placed on that board somehow, to be automatically selected by pick and place machines or manually by human operators. Each component will have a part number, shape code, orientation etc.
A first step in using AOI for process improvement would be to combine placement data with the AOI inspection data.
Like any tool that you might employ to improve your process, AOI can be used in two ways. It is up to you as a user to decide which path or combination of paths offers the best return for the effort required to implement. The steps that can be taken mirror parts of the Six Sigma process.
Firstly the process engineer would identify defects and repair those defects. The Post Reflow Defect Pareto should be known and it need not be limited to an AOI Post Reflow Defect Pareto but can also come from AXI, ICT or manual inspection. Once the Defect Pareto is available the information can then used to determine the right Defect Prevention Strategy:
Post-reflow
Pre-Reflow,
Mixed Mode, or
3D Paste
In step 2 the process engineer can use the Defect Prevention AOI to collect both attribute and measurement data to trace the Root cause of the Defects.
In step 3 manual Closed Loop can be implemented to reduce/eliminate root cause of defects. Once special cause variation has been reduced or eliminated the process engineer may want to move on to step 4.
Finally in step 4 Measurement Analysis is implemented to monitor performance of the placement machine and detect early deterioration in the machine performance before it can lead to the generation of defects. The real goal of Defect Prevention is for all this to happen without human intervention; where AOI data is directly feedback to the Placement Machine, which then uses the data to automatically self correct a drifting process.
Making data accessible means distilling the inspection and measurement data produced into chunks that are relevant and easily understood. For instance a process engineer may be gearing up a new product. The first step in process control is to identify the defects and do some root cause analysis; an information system therefore needs to present data on the production defects detected and where possible an initial identification of how and where the defect was produced. The data should be live (update after each assembly inspected) and any charts and tables displayed to the user should be updated accordingly after each inspection. The accuracy of this data is crucial; no point fixing something that is not broken. So for data to be believable it must be accurate and relevant.
To allow the user to react to defect situations quickly, any software tool must highlight the parts of the production line that are causing the most problems. This allows the process engineer and his team to address and fix these first and move on to the next most important defect cause. The draw back with adopting a find and fix approach is that defects have to be produced to trigger corrective actions. Depending on how many defects are produced on the production line this may or may not be a problem for the producer where ultimately, shipped product quality is how their performance will be measured.
When the production of the new product becomes stable (i.e. common cause defects are known) the type of data presented changes. Continuous process improvement takes over and measurement data becomes important. Again the process engineer could drown in a sea of measurements. Looking at individual component measurements would not help an engineer solve any problems. Out of control conditions should be detected by grouping measurement data according to the section of the manufacturing line that produced each part of the product. Moving to defect prevention means that the SMT line is in control (i.e. all of the common cause variation is known). Unless the line is in control then measurement data will have no meaning or will be misleading at worst.
But you have to be prepared to implement defect prevention. When detecting defects the inspection machine you use does not have any special requirements, other than the inspection algorithms it employs should have sufficient coverage to catch the defects known to be produced by the SMT line. With defect prevention the AOI machine has to be measurement capable. A capable measurement tool typically is defined as:
Capable is <10% gauge repeatability and reproducibility performance (GR&R).
Measurement is the ability to gather X, Y, and Theta offset for every component inspected.
Delivering a usable AOI closed loop solution to the production line can be accomplished in many ways but to do so by efficiently using the data available needs some specialist software tools.
A single defective product delivered to an end user can signal trouble. Recently the author encountered a case in the automotive industry where a single PPM defect returned by their customer forced a process change. This is a costly exercise and would have been better solved by trapping or predicting the defect before it left the production facility. Information systems coupled with real time inspection and measurement machines and clever data analysis software are a necessary requirement for electronics manufacturing where 'Easily accessible and accurate data is everything'.
Peter Conlon works for Agilent Technologies.
Contact Details and Archive...
Related Articles...
Most Viewed Articles...