Acceptance Sampling

ISO 9001 section 8.2.4 says that the organization must monitor and measure the characteristics of the product in order to verify that the product meets requirements, but is silent on the methods to be applied to monitor and measure product characteristics.  The choice of inspection methods is left to the organization. 

Acceptance sampling is one such method.  Since it is not practical to inspect every item in a large batch, acceptance sampling allows you to infer batch quality by examining a random sample from the batch.  Statistically designed acceptance sampling plans, as a method of measuring product conformity, have been around since at least World War II.  Sampling plans like MIL-STD-105, authored by Harold F Dodge and others, have been in use for over 60 years.

Because not all items in the batch are examined, there are risks associated with statistical acceptance sampling.  Two risks are typically calculated:

·         The Acceptable Quality Level (AQL), also known as producer’s risk, is the percent defective that is likely to be accepted 95 % of the time.  There is a 5% chance that product of higher quality than the AQL will be rejected by the sampling plan.

·         The Lot Tolerance Percent Defective (LTPD), also known as consumer’s risk, is the percent defective that is likely to be accepted 10% of the time.  There is a 10% chance product as defective as the LTPD will be accepted by the sampling plan.

Each sampling plan has an associated operating characteristic curve (OC) which describes the probability of lot acceptance as a function of the lot’s percent defective.  The AQL and LTPD represent two points on the OC curve.  As the lot percent defective increases, the probability of accepting the lot based on the sampling plan that the OC curve represents decreases. 
In addition to the risks associated with acceptance sampling plans, there are some practical disadvantages:

·         While acceptance sampling greatly reduces the number of items inspected, other sampling methods such as statistical process control reduce inspection even further and provide process control feedback.

·         When a lot is rejected, we will know why the lot is rejected, but we will not know the root cause of the defect.  We will only know that the product is defective.  There is no control mechanism that will help us control the process the product comes from.

·         Acceptance sampling assumes random sampling, but in most cases the sample is stratified because the product is normally stored in boxes.  As such, it is possible that some product will never have a chance of being sampled and the sample will not be random.

·         Sampling plans based on AQL, LTPD, or AOQL (Average Outgoing Quality Limit) assume that a certain amount of defective material is acceptable.  This sends a message to employees and suppliers that some level of defectiveness is acceptable.  This is not the best message to send if the organization is trying to be a best in class producer.

Sampling plans can be designed by users, or selected from standards such as MIL-STD-105, now obsolete, or ANSI Z1.4 which implements the same plans.  A common pitfall suffered by users not familiar with sampling plan design is to design constant percentage sampling plans.  Avoid constant percentage sampling plans (a fixed percentage of the lot is sampled regardless of lot size).  For small lots, constant percentage sampling plans may not afford enough protection.  For large lots, an excessive amount of inspection will usually result, and the sampling plan will be over critical.

Consider a sampling plan that samples 10% of a lot.  For a 50 piece lot, a 5 piece sample will result in approximately a 1% AQL, but a 46% LTPD (a very weak sampling plan).  For a 5000 piece lot, a 500 piece sample will result in an AQL of .001% and the LTPD will be .46% - a plan unlikely to accept any lot.

While there are better ways to control production processes than acceptance sampling, acceptance sampling can be an effective method for a customer to protect itself from accepting defective purchased product.  Since the customer has no control over the manufacturing process, it is not important for it to understand what process variable caused defective material.  It need only know that the product is defective.  The customer is susceptible to stratified sampling, but accepts this disadvantage in favor of inspecting a small percentage of the entire batch.

For more information see www.rosehillsystems.com

 

Process Control using Shewhart Control Charts

ISO 9001 section 7.5.2 talks about controlling processes.  Some processes are controlled because the quality of the resulting product cannot be determined once the process is complete as 7.5.2 indicates, but there are many other good reasons to perform process control.  Collecting data on a process for the sake of collecting data is of little value.  It is best to collect and analyze process data. 

There are many tools for controlling processes.  The one I’ll talk about today is statistical process control using Shewhart Control Charts.  Walter Shewhart developed control charts around the turn of the last century.  They are tried and true, and I have used them to control a number of processes.

Control charts are based on the theory that the mean of a sample from a normal distribution is normally distributed, with the mean equal to the distribution mean and a standard deviation equal to the distribution standard deviation divided by the square root of the sample size.  In fact, for distributions that are non-normally distributed, the mean tends to be normally distributed.  So the technique is robust.

Shewhart control charts control the process mean by controlling the sample mean.  They control the process standard deviation by controlling the sample range.   Control is applied by calculating control limits for the sample mean and sample range and plotting the sample mean and range on mean (XBAR) and range (R) charts respectively. 

When a mean or range exceeds its control limit, we say that the process has gone out of control.  A data point that exceeds a control limit causes us to stop the process to understand the root cause of the out of control condition, or, more likely we make a process adjustment and note the adjustment on the chart.  For example, it is common for a machine tool to go out of control due to tool wear.  A simple machine adjustment brings the machine tool back into control.

One complaint about Shewhart Control Charts is that they are complicated to set up.  Some also complain that they require the machine operator to stop and make calculations.  In the old days, calculators and computers were not available, and it was more work for the machine operator to calculate the mean and range.  In the digital age though, these complaints are unfounded.  Software to create control charts is easy and inexpensive to come by, and some inspection tools actually do the work of calculating and plotting the control charts.

I put together a simple Excel spreadsheet that will do all the work for you.  You can download it here.  The spreadsheet has some fictitious data placed there so you can see the charts in action.  You can easily overwrite it with actual data.  Instructions for creating a control chart using the spreadsheet can be found here.  Instructions for using the control chart spreadsheet are also provided, and you can download them here.

For questions or help go to www.rosehillsystems.com

Process Control and the Process FMEA

ISO 9001 section 7.5.2 addresses the control of production processes where the resulting output cannot be verified by subsequent monitoring or measurement.   In these cases, defects only become apparent during the use of the product.  Examples of processes that fall within this section are:

• Welding
• Brazing
• Wave Soldering

There are many others.

Control of processes like these is accomplished by developing a well-defined process, validating that the process is effective, and controlling process variables using techniques such as statistical process control.

One tool for validating that the process is effective is the process FMEA (pFMEA).   Process FMEAs are interesting because they have wide applicability to a broad range of processes beyond those covered by 7.5.2.

FMEA stands for Failure Modes and Effects Analysis.  The technique has been around for a long time, having been first developed by the US Military (MIL-P-1629).  Many industries, such as automotive and aerospace have embraced the FMEA approach to both design and process validation.

In performing a process FMEA, the various steps in the process are presented in a spreadsheet.  Each process step has some likely failure mode(s).  Each failure mode has one or more root cause(s). 

For each root cause there is some effect on the product.  This is known as a failure effect.  To each failure effect is assigned a value for Severity (1-10).  The root causes of the failure effect are assigned frequency of occurrence (1-10), and a likelihood of detection (10-1). These three values are multiplied together to come up with a single Risk Priority Number (RPN).

The RPN is used to prioritize actions to improve the process.  As action is taken and process steps are modified, the RPN of the modified process is calculated to show the relative improvement.

There are 10 steps to conducting a pFMEA:

1. Review the process.  Create a process flow chart.  Liest each process step in a pFMEA spreadsheet.
2. Brainstorm the potential failure modes of each process step.
3. List the potential failure effects of each failure mode.
4. To each failure effect assign Severity (S) rankings (1 means not very severe, 10 means very severe).  Record the highest value among the failure effects identified as the severity ranking for the process step.
5. Identify the potential root causes of each failure mode.  To each root cause assign Occurence rankings (1 means occurence s are rare, 10 means the root cause occurs frequently).  When there is more than one root cause, reaord the highest value among the root causes.
6. To each root cause assign a Detection (D) ranking (10 means the root cause is unlikely to be dteected, 1 means it is very likely to be detected).  When there is more than one root cause, record the highest value among the root causes.
7. Calculate the RPN = S x O x D
8. Develop an action plan to improve the process, prioritizing on the highest RPNs.
9. Take the actions in the plan.
10. Recalculate the resulting RPNs after the actions are taken.

The FMEA method is not perfect.   The values assigned for Severity, Occurrence and Detection are somewhat arbitrary.  Moreover the values assigned are ordinal numbers or rankings, meaning a number assigned to a level.  They could as easily represent descriptions such as very low, low, medium, etc.  A higher value for a factor is more important than a lower value. 

Multiplication of ordinal numbers is not defined (We can’t come up with a numeric value for ”Low” x “High”).  A value of “4” is not necessarily four times as important as “1”.  The value “4” is just more important than “1”.  Still, the RPN has value, and when used properly the FMEA can direct a team towards process improvements.   In addition, the completed pFMEA provides evidence of how the process was developed.

A spreadsheet that can be used to perform a pFMEA can be downloaded here.  It is based on the process FMEA promoted by the Automotive Information Action Group (AIAG).  An example of a simple FMEA can be found here.   

For questions or assistance with FMEAs go to www.rosehillsystems.com

Customer Satisfaction

ISO 9001 section 8.2.1 states:
"As one of the measurements of the performance of the quality management system, the organization shall monitor information relating to customer perception as to whether the organization has met customer requirements.  The mehtods for obtaining and using this information shall be determined."

Measuring customer satisfaction means understanding what your customers think.  There are many ways to glean this information including:

• Customer data such as quality and on time delivery reports
• Customer compliments or complaints
• Warranty claims
• Customer satisfaction surveys

The problem with the first method is that most customers don’t provide them, so you’ll only hear from a few customers and at most monthly.  It’s hard to generalize from just a few customer reports.

Customer compliments are rarer still.  It’s more common to hear from a customer when something goes wrong than when you’ve done something right.  Complaints tell what’s going wrong, but do not reflect the good the organization is doing.

Warranty claims are a good way to tell what goes wrong and how much it’s costing.   With this type of data, preventive actions can be focused and feedback on the effectiveness of actions taken will show up in a short time.

Perhaps the best way to monitor customer satisfaction is to ask your customers what they think.  Toward this end, customer satisfaction surveys are a preferred method of listening to customers.  There are many ways to implement these:

• Bingo cards - A bingo card is a small post card on which you ask a few questions and allow the customer to rate each question on a scale like 1 – 5. The other side has your address.They are typically placed in packaging are perhaps the least effective survey method. Few customers complete and return them.The card usually goes to the wrong department.A receiving department will probably just throw them out. It’s unlikely that they will get to the end user, and when they do, the end user is unlikely to return them. I tried these for several years and got less than a 1% return.
•  On line surveys from web sites like SurveyMonkey.com can be helpful, but again, the customer must be motivated to go to the site and complete the survey. Some companies run contests and award prizes as an enticement. Some retailers even offer discounts on future purchases for completing the survey. They get the data they want, and possibly produce another sale.    
• Salesmen’s feedback – One approach is to require the sales force to contact their customers periodically and ask a few well designed questions. The salesman completes a survey form and turns it in for analysis. This can be a good source of data, but the customer will be reluctant to indicate that the salesman is not meeting their expectations.
• The best method I’ve seen is the telemarketing survey. Telemarketers, properly trained, can get to the end user, ask a short list of well-designed questions, and get good feedback which can be analyzed and studied. At the same time, the telemarketer will probe the customer for other potential business. While costly, a well designed and implemented telemarketing survey will reward the company with excellent data, and additional sales that may exceed the cost of the survey.

Whatever method you choose, keep in mind that you should be able to analyze  the data and draw conclusions.  For help with this topic go to www.rosehillsystems.com

 

        

 

Analysis of Data

ISO 9001 requires organizations to analyze data to demonstrate the suitability and effectiveness of the QMS, and to indicate where improvements can be made.  There are many sources of data to analyze, and the organization is at liberty to decide what data to study.

Typical data analysis might include these types of data:

• Outputs from manufacturing processes, such as inspection results, warranty claims, etc.
• Outputs from the corrective action process
• Business process data such as on time delivery, customer satisfaction surveys, etc.
• Vendor performance data such as on time delivery and quality performance

Data analysis can be complex, requiring sophisticated computer programs, or can be quite simple.  Pareto analysis, which separates the critical few from the trivial many, is a simple but powerful technique for determining where to focus improvement efforts.  This simple technique is best shown in a bar chart.  I like it because the results are clear and easily understood by senior managers. 

Information from statistical process control charts such as Shewhart control charts and many variants can be a powerful tool for pointing out processes in need of improvement, and for bringing manufacturing processes into statistical control.  It is not difficult to create spreadsheets that perform these analyses.

Trend charts can indicate those processes which are moving in an unwanted direction.  For example, charting warranty claims monthly can be an indicator that quality of product in the field is deteriorating.

More complicated analytical techniques such as analysis of variance, regression analysis, and designed experiments are useful for understanding the effect of process variables on process outputs.  These analyses can sometimes be done in a spreadsheet, but most often, more sophisticated software is employed.   They require a trained statistical analyst.  Most often, they are applied when troubleshooting process problems.

Organizations should identify those measurements that are important indicators of QMS performance.  Typically, data that is in alignment with the organizations quality objectives will be measured against established goals. 

The output of the data analysis processes can become the input to the preventive action and management review processes.  Using measurement data in this way keeps the organization in line with its objectives and allows it to adjust the QMS as necessary to meet the objectives that have been set.  When selecting measurements, and measurement methods, keep in mind the organization’s established objectives and the audience that will view the results. 

For more information or help go to www.rosehillsystems.com

Quality Costs

In 1980, I bought a 1980 Pontiac Phoenix.  It was the worst car I ever owned. 

In those days I had a young family, and did minor auto maintenance such as oil changes myself to save money.   But I never changed the oil on this vehicle.  It seemed that every time the car was due for an oil change, it was at the dealer for some sort of repair, so I’d just have the dealer change the oil.

I was so disappointed with the vehicle, that I swore I’d never buy another GM car.  That was 32 years ago, and I’ve kept my promise.   My next car was a Japanese car.  I liked it so much I’ve been buying them almost exclusively ever since.  Can any company afford customers like me?

 Many believe that the cost of a field failure is the processing cost of the repair.  Genichi Taguchi, a well-known Japanese statistician and engineer, once theorized that the cost of a field failure is ten times the cost of the product.  If he is right, companies should spend most of its quality budget on prevention.

Some American managers believe that producing a quality product is too expensive.  I argue that producing a quality product reduces cost.  For a real life example, click here.  The problem is that most companies don’t really know how to measure quality cost. 

In many companies the major component of quality cost is appraisal (inspection).  Management knows what that cost is because it’s easy to measure.  They know how many inspectors there are and the associated overhead. 

There are other costs associated with quality though.   In general there are four areas of quality cost to consider:

• Prevention
• Appraisal (Inspection)
• Internal Failures (scrap & rework)
• External Failures (warranty claims, recalls, customer attitude)

In fact, all but the last cost are relatively easy to measure.  External failure cost or warranty cost is more elusive.  We can measure the cost of the return process – shipping cost, evaluation cost, and repair cost, but we can’t measure the customer’s attitude towards the failure.   Will he purchase the product again, or will he buy from someone else?  Will he share his disappointment with friends and colleagues?  I’ve shared the story above with hundreds of people.  Did any of them select a competitor’s automobile as a result?   Who knows?  Can any manufacturer afford to try to find out?

When considering where to spend a limited quality budget, invest in measurement and prevention.  Measurement indicates the problem areas.  Prevention works in the problem areas to reduce or eliminate them.  As the problems diminish, the costs of appraisal, internal and external failures go down.  The result: higher quality and lower cost.

For more information go to www.rosehillsystems.com

Identification and Traceability

ISO 9001 section 7.5.3 addresses identification and traceability of product.  Identification is normally performed by assigning a part number.  Purchased components normally contain a part number from the supplier.  When this is not the case, one should be assigned to the parts or to them upon receipt.

During production, a product that goes through multiple process steps must be identified as to which process steps it has gone through.  This is often accomplished by attaching a traveler document.  A traveler will indicate which process steps have been accomplished.  This is critical when the process step does not change the appearance of the product like passivation, cleaning, inspection and annealing.   Another way to do this is by changing the part number as the product goes through different process steps.

Position can be used as an indicator of production or inspection status.  For example, purchased product is not moved to inventory until inspection has occurred.  Any product not in inventory is considered uninspected.  All products in inventory are considered inspected.  Once removed from inventory, the product cannot be returned to inventory until it has been inspected.

If a product is removed from inventory, processed and returned to inventory, its part number should change when it is returned to inventory.  For example, an item that is plated and returned to inventory awaiting additional processing should have a different part number after plating.

Some parts cannot be marked with a part number.  In this case, the outer packaging should identify the part number of the item, and the item should be left in this packaging until consumed.  In cases where this is not possible, such as items that go into a clean room, the product should be placed in some acceptable tote container, and the container should be marked with the part number.

Test status must also be maintained in all stages of production.  Again, a traveler can indicate inspection status, or an inspection report can be attached to the product.  Also inspection steps that release only inspected product to the next process can be an indicator of inspection status, but is a less secure method.

While identification relates to part number or status, traceability relates to specific parts or production batches.  Traceability is required in some industries.  Lot control is easy to implement and should be considered by most organizations.  Lots can be controlled by marking them by production date.  In many cases, traceability is lost when the packaging is discarded.  In these cases, shipping paperwork should record which production batch (es) were shipped in each shipment.

Some see traceability as unnecessary extra work, but an ounce of prevention is worth a pound of cure.  Consider a product defect that gets into the field.  Knowing the production date where the defect started provides a method of creating a containment plan.  Knowing where that product was shipped, permits a positive recall that assures that only the potentially defective product is recalled.

In one company, we maintained serial number control of finished product and recorded lot numbers of each component on our inventory pick tickets as components went into assembly.  We had received a substandard batch of 14 bearings that started failing in the field.  We were able to recall a select group of 20 serial numbers (there were two lots of bearings in the production batch) for re-inspection and bearing replacement where needed.  Only a few customers needed notification, minimizing our exposure.  We were also able to identify the supplier and charge back the supplier for the defects.

Traceability is important when multiple suppliers supply the same product.  It can be a powerful tool for defect reduction and quality improvement.  Use it wherever possible.

For more information go to www.rosehillsystems.com