A few weeks ago, I wrote about the best way to get your manufacturing automation timeline right. And one conclusion was that a factory should develop its maintenance capabilities before jumping into full automation.
The question is, how do preventive and predictive maintenance work, and what is the logic for putting them in place in your factory?
The Need For Preventive Maintenance
For most production equipment installed in a typical factory, it makes sense NOT to wait for that equipment to break down. There are two reasons for that.
Proper preventive maintenance (regular cleaning, lubrication, adjusting…) makes your processes better able to produce consistently within specifications – in order words, with higher quality.
Another advantage is lower costs. Here are typical numbers:
- Unplanned repair (once machine is down because, for example, a cutting tool is broken): 150 hours of down time, 130,000 RMB of parts, labor, and expedited shipping (because the tool might also have broken another component, because it created quality issues before it broke, and so on).
- Planned repair: 10 hours down time, 20,000 RMB of parts and labor.
How to set the right type of preventive plan?
Let’s say your factory has 20 pieces of a certain type of equipment – for example CNC milling machines, or plastic injection presses. Where to start?
Here is a set of rules that makes a lot of sense.
Source: Maintenance, Replacement and Reliability: Theory & Applications, 2nd edition, by Jardine and Tsang, CRC Press.
3 Common Predictive Monitoring Approaches
By gathering past break down data and running a statistical analysis, it is possible to predict the risk of failure for each piece of equipment at different times in the future. Let’s look at the three most common approaches.
1. Condition monitoring
This approach entails:
- Setting an inspection schedule (e.g. recording the temperature, vibrations, and noise or a bearing);
- Planning for certain actions if the findings are beyond a certain threshold;
- Using the results from those inspections to predict the timing of the next inspection.
The great thing about this approach is, it gives very granular directives (at the level of each individual piece of equipment).
Why is this important? You might have 20 machines, but do they all have the same hazard rate? Maybe not. They don’t process the same parts, they are set at different speeds, they had different incidents in the past, and so forth. Making predictions as if they were a homogeneous group might not make sense.
2. Time-based maintenance/discard (for similar assets as a group)
Preventive maintenance usually takes the form of time-based maintenance (e.g. “lubricate every 30 hours of operation”), but it is often based on guesswork. Injecting some predictive power will improve the choice of the time intervals.
Let’s take an example. Let’s say there is a 15% chance of failure on the CNC milling machines 20 days after they were last maintained. Knowing this, the right actions can be taken – each machine can be stopped for inspection when they reach that milestone, spare parts can be ordered a little in advance, etc.
Statisticians have done an amazing job of developing formulas that account for very incomplete data (as is often the case in Chinese factories). Even if you only have past failure incidents data on 5 or 6 of these 20 machines, you can still rank them and run a Weibull analysis.
3. Can time-based maintenance/discard be done for each individual machine?
Yes. And this is often a welcome refinement.
A Weibull analysis can be run based on the past failure data of each piece of equipment. If you have these data about some of your machines, you will be better equipped to forecast each machine’s mean time to failure and other useful statistics.
I hope these rules of thumb about setting up a preventive and predictive maintenance system have been useful. Let's just recap:
- Predictive maintenance can make your preventive efforts much more time- and cost-efficient with the use of a statistical software package such as Minitab.
- Before analyzing data about all your machines as if they were a homogeneous group, check if condition monitoring is feasible.
- If it is not, set the right time intervals for time-based interventions.
What do you think? Have you been applying some of these tools? What results? What hasn’t worked as well as you hoped? Leave a comment below and we will make sure to respond.