Defect Generation and Detection
The quality level seen by the customer of our process is one of the most critical process metrics. This quality level is the result of all the defects generated by the different process steps as well as those coming from the parts suppliers. In our Value Stream Map we focus on process flow but this flow is very much affected by quality. It is important to find out where in the process each type of defect is generated and where in the process it will be detected to make sure it doesn’t find its way to the customer.
Download this Excel simulator DPUdpmo.xls from OneDrive folder:
In this model we are simulating a simplified map of an Electronic Assembly and Test Process.
Defect contributors are defective components coming from suppliers and defects from the 4 process steps:
- SMT (Surface Mount Technology) automatic component placement
- Reflow solder of these components
- Manual PTH (Pin Thru Hole) component insertion
- Wave soldering of PTH components.
Defects are detected on:
- Several visual inspections along the process
- ICT (In-Circuit Test) at the end of the assembly process
- Final Functional test.
“Defects per unit” (DPU) is the metric that interests the customer but it’s not very practical as a process metric because we are typically producing products with very different levels of complexity: some boards may have just 5 components and others 500.
“Defects Per Million Opportunities” (DPMO) is a useful process metric that is independent of the product complexity. This metric will allow us, for instance, to know if our soldering process is improving or getting worse with independence of the complexity of the products we are soldering.
In the following example: if we have one defective component in 3 boards the average DPU = 1/3 = 0.33
When we measure in DPMO we must define the OFEs (Opportunities For Error) of the process we are dealing with:
- Component placement process: OFE = Number of components placed
- Soldering process: OFE = Number of solder joints soldered
Therefore if we say our component placement DPMO = 100 this means that out of every million components placed 100 will be missing or misplaced.
If our wave solder DPMO =200 this means that out of every million solder joints 200 will be defective.
In our Electronic Assembly example we have 500 SMT components and 100 PTH therefore these are the corresponding OFEs. Total components used por board are therefore 600.
The 500 SMT components have a total of 1500 solder joints, therefore these are the reflow soder OFEs.
The 100 PTH componntes have a total of 300 solder joints (wave solder OFEs).
Process DPMOs define the quality of our different process steps and the OFEs are specific for each product type.
On each step we can estimate the expected defects per unit:
DPU = DPMO x OFE / 1 000 000
Since the types of defects contributed by the 5 different sources are independent, the total number of defects produced will be sum of each defect contribution. Total DPU is the sum of all DPUs.
Process Yield is the proportion of defect free units. If defects are independent DPU follows a Poisson distribution so Yield and DPU are related by:
Yield = exp(-DPU)
For instance an average of 1 defect per unit will still produce exp(-1) = 0.37 that is 37% defect free units (there will be units with more than one defect)
Total DPU is therefore the sum of DPUs while the total Yield is the product of yields.
Now let’s look at defect detection:
Control points (visual inspections and tests) are defined in the Control Plan for each of the product types. In this example we have several visual inspections, one ICT and one Functional Test performed in this order.
Visual inspection is performed as close as possible to the step generating the defect and the defects detected are typically corrected on the spot. The problem is that it is very inefficient: it can only catch 60% of the defects. In this example this means that out of the total 0.32 DPUs only 0.192 will be caught and corrected, therefore 0.128 will escape.
ICT uses a bed of nails to contact many points in the board and is able to test the functionality of some components. If the test coverage is 80%, 0.102 DPU will be caught in ICT and corrected and therefore 0.026 will escape.
Finally the functional test has a high test coverage but the test is not able to pinpoint where the defect is, therefore faulty boards need to be analysed and repaired by an expert through a lengthy process.
The proportion of boards that will need to be reworked as a result of visual inspection is 17%. This will require additional resources and skills. In the case of ICT and functional test 12% of boards will require rework and this may require a higher level of skills and it will take longer to identify and correct these defects.
At the end of this process we have 0.001 DPU going to the customer which means 1280 faulty boards out of every million.
This is obviously unacceptable so we can start looking at the main contributors to defects and the main escapes of our detection and correction.
We can look at different improvement alternatives with this simulator and estimate the effect on the final ppms going to the customer.
PTH insertion, being a manual operation, is the main contributor to defects. We can look at ways to add some Poka-Yokes (defect proof devices) or look at ways to automate. Suppose we are able the reduce the PTH insertion DPMO from 1500 to 500 the result would be a reduction of output to 1120 ppm.
If we can further improve visual inspection coverage by automation to 80% them it would go down to 560 ppm.
If we redesign the board to eliminate all PTH components and convert them to SMT we would further reduce it to 384 ppm.
In this simulator Process DPMOs come from measuring our current processes. Theoretically they should be Product independent but in actual fact they are not: some designs produce more manufacturing defects than others and the process parameters for some products may not be optimized.
This simulation helps us focus on the main contributors of defects and defect escapes from our tests. In this way we can identify who has to do what in order to improve the process.