Towards the development of a maintenance concept: how to deal with uncertainties – Lely
|Company Name / Department||
Lely – Oscar Moers
TU/e – Claudia Fecarotti
TU/e – Néomie Raassens
|Housing arranged by company||–|
400-500 per month
Working at Lely
It is an amazing opportunity at an international innovative company, leader in the agricultural industry. In other words, enough to learn for an intern!
- Working in an international environment where you can really make an impact with your contribution;
- You will work at one of the most innovative organizations in the Netherlands;
- Freedom in organizing your own work;
- Lots of responsibility;
And the best cappuccinos made by our own barista and fresh milk directly from our farm from one of our colleagues.
Below are some useful links:
Lely is traditionally an Original Equipment Manufacturer (OEM) which supplies innovative products along with some technical service including necessary repairs and warranty. Nowadays Lely is undertaking a “servitization journey” aimed at upgrading the company from an OEM, to a service provider. This means that Lely will not only supply products, but the technical services provided to the client as per contract will be a tailored preventive maintenance service optionally bundled with a lump sum for break-down services. The offered maintenance services develop from corrective maintenance, via preventive and predictive maintenance to pro-active maintenance. One of the goals of the servitization journey is to offer customers 100% uptime for Lely products, with no unscheduled breakdowns and limited number of scheduled service visits, along with a minimization of the maintenance costs and a maximization of the product output performance. The offered maintenance services develop from corrective maintenance, via preventive and predictive maintenance to pro-active maintenance.
This research plan is meant to contribute to Lely’s servitization journey by ultimately developing a decision support system for initiating maintenance actions and their clustering for a machine as a whole during the entire life of the machine. The decision support system will develop “optimal” maintenance concepts for the machine during the (1) design phase, (2) early exploitation phase and (3) full exploitation phase, respectively. Each phase has different requirements and challenges related to the uncertainty of the failure and degradation processes, which are strictly dependent on the availability of engineering and field data. The decision support system should work as one system, but embed models tailored for each phase.
The ultimate vision is to have a machine with maintenance concept that is customer specific optimized on cost, downtime and performance, by fully implementing predictive and pro-active maintenance. The relevant components are continuously monitored by means of sensors. The IoT (Internet of Things) technology combined with AI (Artificial Intelligence) techniques will enable the timely prediction of failures and times to degrade to relevant degradation thresholds, as well as the selection of the appropriate maintenance action. Accurate predictions of the remaining useful life of components will enable to minimize the loss of life while also minimizing the risk of unexpected failures. The decision support system will enable “individualization” of the maintenance concepts based on customer preference, external circumstances, service conditions and machine usage.
Maintenance models focused on optimizing maintenance policies and maintenance concepts, usually rely on the assumption of a lifetime distribution and a degradation process and corresponding parameters, for components subjected to age-based and condition-based policies respectively. However, estimates of lifetime distributions and stochastic degradation processes are often affected by a high level of uncertainty (either on the parameters only, or on the model itself). The level of uncertainty clearly depends on the quantity and quality of historical data and online data if available. In practice however, we often deal with low quality and incomplete data (right-censored failure data because of the implementation of conservative preventive policies). The operating conditions also affect the way components degrade and fail. Lely can rely on both field service data and sensor data, however these data are not always reliable.
We consider optimization of maintenance concept under uncertainty of the parameters of the lifetime distributions and degradation processes. As a consequence, the expected cost resulting from a given maintenance concept is also a random variable, and one wants to minimize not only its expected value, but its variance as well. Different maintenance concepts result in different expected cost and variance, and the choice will depend on the attitude of the decision maker towards risk, which can be modelled either within the objective function or as a constraint. We want to calculate the variances of the expected costs (and downtime) given the estimators of the uncertain parameters and their variances based on observed data.
Goals of the Project
Answers to the following questions:
Q1. How do we estimate the uncertain parameters of the lifetime distributions and degradation processes? The estimation method is strictly related to the “size” of the sample data. Possible methods range from classical statistical inference for large sample size (e.g. maximum likelihood methods) to expert elicitation and Bayesian inference for smaller sample sizes.
Q2. How do we calculate the estimator of the maintenance costs and downtime and their variances? How do we calculate the bias between the “true” maintenance concept and related costs and downtime, and their estimates?
The expected maintenance costs and downtime are functions of the uncertain parameters. The suitability of methods to be used to evaluate the cost and downtime estimators based on the estimators of the parameters, depends on the size of the sample data and the fulfillment of some properties (e.g. for using the traditional delta method, one has to prove the asymptotic normality of the estimators, both the parameters’ and the optimal cost/downtime functions’ and maintenance policies’).
Q3. How does the parameters’ uncertainty affect the maintenance concept and related system’s performance? What is an “acceptable level of uncertainty”?
A sensitivity analysis of the maintenance concept to the parameters is carried out. Here we can also run the analysis for different lifetime distributions. Confidence intervals for the “true” optimal cost and downtime can be found and the way these intervals are influenced by the uncertain parameters.
Q4. How can we improve our maintenance decision making while accounting for parameters’ uncertainty due to poor quality/quantity data?
Q5. Implementation to Lely Discovery Collector 120.
Essential Student Knowledge
More information: email@example.com