This blog reviews a best in class CNC machining process validation program. Our author writes, “In general, the best approach is a risk-based approach.”
The original question from a former client was: “What does a best in class CNC machining process validation program look like?” Although I intend to answer this question, I know a few other clients that have done a great job of this. Hopefully, they will add their own opinions as a comment. Therefore, I am expanding the scope of this question to validation in general.
The problem with validation is that you can always do a more thorough validation. Only in the cases of processes, such as sterilization, do we have ISO Standards that tell us what is required. Otherwise, we are usually the experts, and we have to use our judgment as to what is necessary. In general, the best approach is a risk-based approach.
For each design specification established for a component, we also need to identify what process risks are associated with failure to meet the specification. Most companies perform a process Failure Modes and Effects Analysis (pFMEA). This risk analysis has three quantitative components: 1) severity of the failure’s effect, 2) probability of occurrence, and 3) detectability. The first factor, severity, is based upon the intended use of the device and how that component failure impacts that use. Usually, it is important to have a medical professional involved in this portion of the estimation.
The second factor, probability, is typically quantified during process validation activities. One company I audited developed a ranking scale for the probability that was linked directly to the CpK of the process. Higher CpK values received lower scores because the process was less likely to result in an out-of-specification component. Another company I worked for used a six-point logarithmic scale (i.e., – 10e-6 = 1, 10e-5 = 2, 10e-4 = 3, 10e-3 = 4, 10e-2 = 5, and 10e-1 = 6). This logarithmic scale was based on sterilization validation, where a sterility assurance level of 10e-6 is considered “validated.”
The third factor, detectability, is best estimated by using a quantitative scale that is based upon a gauge R&R study or some other method of inspection method validation.
Most companies struggle with the determination of what is acceptable for design risk analysis. However, for process risk analysis, it is usually much easier to quantify the acceptable risk level.
Once you have determined that a process is not acceptable at the current residual risk level, then you must take corrective actions to reduce the risk. The first step to achieve this should be to review the process flow. There are critical control points that can be identified in the process flow. One of those places is at the end of the process at the inspection step in the process.
The inspection step in the process flow affects the detectability of defects. For many automated processes, such as CNC machining, it is not reasonable to perform 100% inspection. Therefore, these processes require validation. Most engineers make the mistake of trying to validate every dimension that is machined. However, only some of the aspects result in device failures. These are the dimensions that are critical to validate. The best practice is to calculate the process capability for meeting each of these critical specifications (i.e., – CpK). A minimum threshold should be established for the CpK (refer back to the process risk analysis for ideas on linking CpK to risk acceptance). Any CpK values below the threshold require a more consistent process. These are the component specifications that should be the focus of process validation efforts.
During a process validation, it is often advisable to perform a Design Of Experiment (DOE) in order to quantify the effects of each process variable. Typically a DOE will evaluate the impact on CpK for each variable at a high, low, and middle value, while other variables are maintained at nominal values. Any variables that appear to have a significant impact on the CpK are candidates for performing an Operational Qualification (OQ). For a machining process, this could include spindle speeds, feed rates, and material hardness. If variation of the variable has little or no impact upon the CpK, then there is probably little benefit to the inclusion of this variable in an OQ.
The output of an OQ validation should be high and low limits for each process variable that will result in a “good” part. Performance Qualification (PQ) validation is the final step of process validation. In the PQ, most companies will conduct three repeat lots at nominal values for the variables. If the OQ is designed well, there is often little added value in the PQ. Therefore, the sample size is typically three lots of 10 samples each. If the OQ validation does not clearly identify safe operating limits for the variables, or the process has the marginal capability (i.e., – a low CpK), then the OQ should be repeated, and an additional DOE may be needed.
Here are a few information resources for those of you that are in “Deviceland”
- Guidelines for the Validation of Chemical Methods for the FDA Foods Program (3/22/2012) – http://www.fda.gov/downloads/ScienceResearch/FieldScience/UCM298730.pdf
- Process Validation: General Principles and Practices (January 2011) – http://www.fda.gov/downloads/Drugs/…/Guidances/UCM070336.pdf
- Guidelines for the Validation of Analytical Methods for the Detection of Microbial Pathogens in Foods (9/8/2011) – http://www.fda.gov/downloads/ScienceResearch/FieldScience/UCM273418.pdf
- CPG Sec. 490.100 Process Validation Requirements for Drug Products and Active Pharmaceutical Ingredients Subject to Pre-Market Approval (3/12/2004) – http://www.fda.gov/ICECI/ComplianceManuals/CompliancePolicyGuidanceManual/ucm074411.htm?utm_campaign=Google2&utm_source=fdaSearch&utm_medium=website&utm_term=validation&utm_content=3
- Q2 (R1) Validation of analytical procedures: text and methodology (June 1995) – http://www.ema.europa.eu/ema/index.jsp?curl=pages/regulation/general/general_content_000431.jsp&mid=WC0b01ac0580029593&jsenabled=true