A question was posed to me this week about why the OREDA failure rate for a solenoid valve was greater than the Failure Modes, Effects, and Diagnostic Analysis (FMEDA) results from exida for a specific manufacturer’s solenoid valve. The person asking the question was clearly challenging the exida FMEDA results. There are several known reasons why this data is different:
1. FMEDA data is for a specific product design, and OREDA data is for a category of products
Any competent failure data analyst, including the experts at exida, must consider the worst design example when publishing failure data for a generic class of devices. Some designs have much greater failure rates than others. Some are much older designs with higher failure rates. FMEDA data can show the differences in these devices, but field failure data collection often lumps these diverse products together in order to have sufficient data samples. The exida generic solenoid valve failure data does the same thing and assumes worst case design based on the exida Predictive Model and compares closely to OREDA. Specific FMEDA results apply to a specific design, and specific results are often much lower as is appropriate to a superior design.
2. Product Specific Failures vs. Site Specific Failures
FMEDA failure data includes only “product specific” failures. Failures due to random maintenance failures are site specific, and are not included in the numbers. During a discussion at a safety symposium in Norway last year, I was told that the OREDA database includes these site specific failures. Therefore, this would result in a higher number. Now I agree that site specific random failures should be included in a PFDavg calculation but at exida, our exSILentia tool models site specific failures separately. We believe this to be a far more accurate method.
Field failure data studies are essential to verify and calibrate other techniques like FMEDA. After reviewing hundreds of field failure studies with an accumulated sixty billion unit operating hours of data, FMEDA results are becoming quite accurate. When combined with a good modeling tool that also accounts for site specific failures, realistic PFDavg calculations can certainly be done.