Other necessary optimization methods

The goal of this research was to come up with a way to gather information about failure behavior. The proposed method(s) to do this was presented in chapter 4. The question is what can be done with this information. As could be seen in the previous sections, one can plan maintenance by some simple formulas. The formulas presented however, are based on an age replacement policy (replace component when a certain age is reached). An also they assume both corrective and preventive maintenance is ‘perfect’. In the sections below, some possible extensions of the simple age replacement model with complete renewal from section 5.3 are discussed briefly.

5.4.1 Block maintenance models

As discussed in chapter 3, there are two basic models: age maintenance and block maintenance models. In an age maintenance model, a component is maintained when it reached a certain age. In a block maintenance policy, a whole block (which would either correspond to a ‘group’ or a ‘cluster’ at Loodswezen) will be maintained at once. Maintenance on a certain component is independent on what happened with the component before. In a very extreme case, a new dynamo for instance could be replaced the day after already in such a policy (this won’t happen a lot though). For this reason, age maintenance policies have a slightly better performance, at least in terms of costs. Block maintenance policies are however way easier to implement as the history of each component does not have to be recorded. In addition, a block maintenance policy could feature benefits in terms of downtime, as one can use shared set-up times more easily.

Loodswezen currently uses an age maintenance policy that is pretty advanced as it is already clustered and harmonized. The main advantage for block maintenance in terms of decreasing downtime is that maintenance activities are put into blocks that are most likely comparable to the current clusters. For this reason, changing to a block maintenance policy would probably not lead to much faster maintenance and lower (expected) downtime.

5.4.2 Partial renewal models & inspections

A more important extension to the model that assumes complete renewal, is models with partial renewal and models that can handle inspections.

When replacing or revising a dynamo, the dynamo is indeed as good as new (complete renewal). When re-attaching the v-belt on the dynamo, the dynamo is functioning again, but still is in a state that is ‘as bad as old’. Sometimes however, maintenance consists of tightening screws or cleaning. This will increase the lifetime of the component, but it will not renew it. The current model assumes that all maintenance makes components as good as new, which is not always the case. To solve this problem, there are many models that assume partial renewal. These models can model every result of maintenance between ‘as bad as old’ and ‘as good as new’.

A very important maintenance type that does not change the state of the system is inspections. After an inspection, the system will not be renewed (unless a failure or near-failure is detected). There are also a lot of models to model inspection. The choice where it is all about here is when to do

inspections as obviously these also cost time. It could be clever to increase the rate of inspections with the age of the components. As a degrading component becomes older, the probability of failure becomes higher and the probability that the component will survive until next inspection will be smaller. There are models that can model the degradation and decrease the interval between inspections when the degradation reaches a certain point. An interesting question for Loodswezen might be what to do with the results of an inspection. Obviously if the oil has a certain amount of contamination, it will be replaced. But when is a dynamo ‘too dirty’, when is a screw ‘too loose’ and when is the alignment of a dynamo not good enough? Even more interesting is the decision to wait with preventive maintenance until next inspection or failure, on what basis is this done?

61 To conclude, in order to model partial renewal and inspections, many models are available. These models can be as simple and as complicated as one wish.

5.4.3 Opportunistic models

For Loodswezen, opportunistic maintenance models are very promising. Opportunistic maintenance is already done at the company, although the decision is currently based on the judgement of the workshop manager. When a component fails and preventive maintenance is nearing (let’s say the engine is at 2800 operating hours and maintenance is planned at 3000 hours), preventive

maintenance might be brought forward. This way, the vessel will be in maintenance only once. This should increase availability.

Of course, the decision is quite hard. If this failure occurs at 2800 hours, is that too early to do 3000 hour maintenance? This decision is currently made by a person but with the use of opportunistic models, one could calculate the optimal threshold to use the opportunity of preventive maintenance. There are a lot of these models with a wide variation in complexity. One has to understand that the decision is somewhat more complicated than “shall we do it now or later?”. Although chapter 4 provides the parameters needed to calculate this, there are more issues involved. For instance inspections that are needed because of warranty. Bringing forward the preventive maintenance now could then lead to the demand from the supplier to bring forward next preventive maintenance too in order to keep the maximum time interval between two maintenance instances the same. This would then lead to no real benefit. In addition, one has to for instance certifications that have to be done before a certain moment and will be valid for, for instance, one year (so you just move the problem to next year). On top of that, maintenance that is done by external companies cannot (always) be part of such a system. In an opportunistic model, the date that maintenance is done can change every day so no appointments can be made with your service suppliers.

5.5

Conclusions

The Weibull distribution that was found in chapter 4 can be used to determine optimal maintenance intervals. In this chapter, a basic age maintenance method with complete renewal is used to

calculate optimal maintenance times. Two ways of giving meaning to these optimal maintenance intervals are proposed:

1. Use existing groups (time intervals) only. 2. Add new groups to the system.

In the dynamo case, the first case means that the dynamo is preventively replaced or revised after 7000 hours, the second case means the replacement is done every 6250 hours. These numbers are based on β = 1,41, α = 667, cost ratio = 3,25 and an engine is used 7,4 hours a day on average. The 7000 hour interval replacement features a reduction in downtime, caused by the dynamo, of 2,1%. The 6250 hour interval replacement gives a reduction of 2,3%. The benefits of the optimization are rather low in this case because:

β is relatively low (aging is relatively slow). The cost ratio is relatively low.

With this, Loodswezen can build a maintenance planning of its own. In order to make this somewhat more precise and to improve performance, block maintenance models, partial renewal models, models with inspections and opportunistic maintenance models can be used.

62