If you just want to meet the intent of the clause, which we do 'cos our process is really stable, so we put our resources elsewhere # : ) we do this:
As with every predictive method, you need to record the data as it happens. This should include the frequency of ALL failures/replacements, the parts that fail/are replaced, the time between each failure/replaced part and the downtime caused by the part failure/replaced part.
Then put these into a pareto chart to help determine either the parts that fail the most or the parts that cause the most downtime or the parts that most costly to replace if they fail.
Then you should determine which of the parts are key parts i.e. you need an action plan in place for when the part fails or needs replacing. This can be done something like the attachment.
Hope they help.
Then decide on what resources you're going to use preventing a recurrence.
I do this every 12 months, so you should have enough information to do this.