You just installed the latest and greatest in predictive analytics and out come the first results. Now what?
Interpretation of the forecast is key. First and foremost it is crucial to understand that no two forecasts mean the same. Let’s take a couple of examples to see how specific forecasts could mean very different things to your operations. Parts A and B are both forecasted to fail with a equal likelihood over the next two weeks. Should I remove these parts? The economic rationale behind this decision is often overlooked when companies set rules like “whenever the failure probability is over xx%, replace the part”.
Let’s look at some of the variables (not exhaustive) which should not be overlooked:
Criticality; are parts A and B equally critical or is one maybe even not critical at all? Take IFE (In Flight Entertainment) for example. For one airline, individual failure may not be critical and they’ll have a rule that up to x (typically one or two) screens may be out of order before the failure causes a no-go situation. For another airline, customer experience may be the number one KPI and therefore IFE screens become AOG parts – one failure and the aircraft’s grounded. In light of this different handling of errors, the 70% indicator may lead to a pre-emptive maintenance or not. Criticality becomes even greater when safety is concerned and in some cases, the slightest hint at something failing may (or should) cause an intervention.
Cost of the intervention: if parts A and B have the same chance of failing but the cost of replacing (or even inspecting) both parts differs greatly, this may lead to different decisions on whether or not to perform the check. Some parts are readily accessible or sit just behind a hatch whereas to access other parts, one may have to dismount part of the asset. Taking this factor into account it is pretty obvious that a 2 minute intervention versus a 2 hour intervention may get different thresholds set as to what confidence level is required from the prediction before action is taken.
Waterfall costs: this is a very interesting one. Some parts may be very inexpensive but failure of that part may cause a bigger asset to stop, or worse, may cause downstream damage. The impact of a 10$ part can thus easily run in the $1,000’s. Failing to capture the full impact by only looking at individual part economics leads to very poor decisions. In this field, input from subject matter experts is key to grasp the total situation. It is very much like putting together a failure framework based on the in-service BOM. The easiest example is a production line; if one machine fails, the whole line comes to a halt and the cost avoided by maintenance (by not intervening) may lead to a huge cost on the production side. Beware of company silos when taking these decisions!
Availability of a replacement part: Like you wouldn’t want to fly on a plane without radar, no decision should be taken blindly. As the previous paragraph has shown about the importance of cross-company information flow, so it is with maintenance. An asset shouldn’t be stopped if the required part (or even the mechanic) is not available and similarly, the predictive analytics side should “warn” the parts department about imminent failure risk in order for them to get the logistics moving to get the part there on time. Systems should take to each other as the outcome of one determines (or at least should) the actions proposed by another.
The bottom line is that, just like supply chain is moving to Integrated Business Design, so should maintenance. The overall economic impact to the company bottom line and the impact to corporate objectives should prime over departmental KPI’s. Now, nobody wants to run a department whose KPI’s get negatively impacted. Therefore, change management and careful objectives design should be part of any move towards predictive maintenance (or any business transformation for that matter). The potential benefits are too great to let them get wasted on corporate politics or underperformance due to poor goal setting.