Q: Our kitting and assembly area recently underwent a major change in work flow. I need to complete a risk assessment audit, but I am unsure of all the areas I should be examining. Is there a tool that will ensure all areas and parameters are audited correctly?
A: A facility risk review (FRR) is a frequently used tool for risk assessment audits. An FRR can provide information that helps focus resources at manufacturing facilities on the most important threats to production, employee or public safety, and the environment. The major objectives of an FRR are:
- To identify the major contributors to the risk of accidents that threaten the safety and health of employees, the general public and the environment, and causes loss of production or capital equipment.
- To develop, where possible, recommendations for reducing risk.
- To identify processes and areas that require more detailed study to determine risk reduction measures.
If applied to a kitting and assembly area, a detailed risk assessment would be a time-consuming and expensive process. An FRR, however, is a much less expensive process that uses the framework of risk assessment to screen areas or processes and ranks them according to their relative risk. The approach requires only order-of-magnitude estimates of the frequencies and consequences of events. It incorporates company-specific experience and industry data to characterize plant risks efficiently.
An FRR first identifies the failures or accidents that can occur at a facility and then determines the approximate consequences of such failures or accidents and their expected frequencies of occurrence. This is accomplished in a systematic manner by dividing the kitting and assembly area into small process or physical sections. Potential accidents will then be postulated for each section; these accidents may involve events resulting from equipment failures, structural failures, human errors and external events.
An FRR characterizes risk by frequency categories and consequence categories, as illustrated in Tables 1 and 2, respectively. The categories in these tables have been arbitrarily selected to illustrate the FRR approach, and other categorization schemes are dealt with later in this answer.
Exact categories are defined for each specific FRR (actual FRR applications usually involve a few more categories than the ones shown in Tables 1 and 2). Each accident of interest considered in an FRR is assigned a consequence category and a frequency category.
For example, in Table 2, an explosion or fire in the kitting and assembly area may result in loss of production for about two weeks (production loss category 2), facility or equipment damage totaling about $5 million (facility or equipment damage category 3), death of an employee (employee safety category 4) and a small release to the environment outside the plant boundary (environmental category 2).
In this example, the accident of interest is not expected to occur during the lifetime of the facility (frequency category 2 in Table 1).
An FRR usually considers many accidents. Once these accidents have been categorized with respect to consequence and frequency, they can be conveniently depicted in a risk matrix, like the example representing facility or equipment damage consequences illustrated in Table 3 (similar matrixes may be developed for other consequences).
The entries in the matrix in Table 3 represent the number of accidents that have been assigned to the frequency and consequence categories associated with that entry. For example, the matrix indicates that 56 accidents have been assigned to frequency category 2 and consequence category 1.
The risk matrix in Table 4 defines regions of low (blue), intermediate (green) and high (red) risk. Low risk requires no further work, but additional analyses and mitigations may be needed for accidents classified as intermediate and high-risk accidents. A significant number of accidents can be eliminated from further consideration in this study, and recommendations can be made for reducing the risk of the accidents that lie in the upper right section of the matrix.
James J. Rooney
Director, quality and lean Six Sigma services
Q: In section A9.1, ANSI Z1.9 discusses the applicability of a mixed variable and attribute sampling plan. Two conditions are given under which a mixed plan can be used. Condition A is if the population is screened to select samples that are within specification. The standard says condition B can be employed if "other conditions exist that warrant the use of a variables-attributes sampling plan." I have not been able to locate a source of information on what those other conditions are. Can you provide any guidance?
Becton, Dickinson and Co.
A: The advantage of using variable sampling instead of attribute sampling is that for the same level of protection, you can use a smaller sample size. Other factors, however, can easily offset any cost savings.
ANSI Z1.9 is not used very much. Attribute sampling is the most popular option by far. The primary reason is because to use variable sampling without qualifications, you must have a normal distribution. Without knowing this, there must be further analysis if the lot is rejected based on the 1.9 standard variable sampling plan and if you have no defectives in your sample.
For example, imagine returning a lot to a supplier because the statistics of the variable sampling plan tell you to reject the lot, even though no actual defects are found in the sample. Assume the supplier has a problem and screens the lot before the supplier sends it to you. In that case, you would receive a truncated distribution. Not aware of the screening, the 1.9 plan used would reject the lot. The normal distribution did not exist. The truncation would be at one or both specification limits.
The second thing to consider is that you can reject a lot that has not been screened, even if there are no defects. In that case, you may not have a normal distribution. There are tests for normality, but the general rule is that if you reject a lot based on 1.9 and have no defects in your sample, go to the attribute sampling plan.
The attribute plans are based on the hypergeometric, Poisson and binomial distributions. The Poisson and binomial distributions are estimates of the hypergeometric distribution. The hypergeometric approach is used by gambling casinos and in the c = 0 sampling attribute plans, while the Poisson and binomial distributions are used in ANSI Z1.4 attribute plans.
While I am sure there are selective applications for Z1.9, in my opinion, it should not be used for general purposes.
Nicholas L. Squeglia