Methods to generate the yearly shutdown-schedule of a basic oxygen steel plant

Bel, Michiel (2013) Methods to generate the yearly shutdown-schedule of a basic oxygen steel plant.

Abstract:Tata Steel’s Basic Oxygen Steel Plant in IJmuiden converts pig iron into sheets of steel, which requires several installations. These installations all require different maintenance jobs, which are specified in the SAP system where the job’s cycle time, duration, priority, start date, etc. are stored. Currently, Tata Steel bases the Yearly Shutdown-Schedule (YSS) too much on the maintenance needs of installations, instead of on the jobs that have to be performed on these installations. Furthermore, Tata Steel generates the YSS based on manual processes, without a formal methodology. Hence, the research question of this research is: Method In the literature, our problem is called the ‘Maintenance Scheduling Problem’ (MSP) and the problem is solved in the literature by applying a local search heuristic, because the problem is too complex to be solved to optimality. Our MSP is even more complex, mainly due to the high amount of possible routes through the Basic Oxygen Steel Plant. In order to create a transparent and subjective method to generate a YSS, we formulated the MSP as a Mixed-Integer Program, which includes the restrictions on shutting down an installation. The method contains four ranked objectives, in which a lower ranked objective can never by improved at the cost of a higher ranked objective: 1. Minimize the overflow of pig iron (deregulating the Blast-Furnaces and dumping at Harsco). 2. Minimize working in overtime (during the weekends and on working days before 7am and after 3pm). 3. Minimize deviating from the job’s cycle time. 4. Maximize the spreading of jobs. In order to solve the MSP, we applied a Simulated Annealing (SA) heuristic that uses four runs to subsequently optimize each objective. Results The analysis of the results of our four-run SA approach clearly shows the correctness of the approach with respect to clustering jobs in order to minimize the overflow: our approach causes a decrease in the objective value of 89% with respect to the initial schedule, as Table 1 shows. In Table 1, the ‘initial’ column shows the values of the four objectives if every job is scheduled at its start date in SAP. Additionally, we applied two Iterative Improvement approaches, a one-run SA approach, and a combined approach to the same MSP. From this analysis, we conclude that both our four-run SA and the one-run SA outperform the Iterative Improvement approaches and the combined approach. Although it did not out- or underperform the method in the small experiment that we performed, we expect our four-run SA approach to outperform the one-run SA approach on the long run, because the one-run SA approach finds a bad local optimum relatively quickly. Furthermore, the analysis shows that not all rules of thumb that Tata Steel currently applies to generate the YSS remain valid from an overflow point of view, whereas we are unable to conclude on the validity of the rules of thumb from any other point of view, such as safety. Although not all of these rules of thumb remain valid, the current YSS seems to outperform our YSS. However, 55% of the jobs do not fit to the shutdowns in the YSS as Tata Steel generated it. This is mainly due to our usage of tight bounds on the allowed deviation from the cycle time: we expect that our approach outperforms the current manual approach if it is based on the same restrictions. Conclusions and Recommendations As mentioned in the previous section, the currently applied restrictions differ from the restrictions as we applied them, which is mainly due to the incorrectness or unavailability of the job’s duration, the priority, and cycle time. We propose determining the correct parameters for every job in SAP, such that the YSS is based on these data, which improves the objectivity of the restrictions and the trust in the output. Furthermore, we recommend clustering jobs in order to decrease the runtime of the SA approach. Especially clustering the jobs on the casting installations decreases the problem size without decreasing the quality of the method and the YSS. Finally, the main advantages of our method over the current method to generate the YSS are mainly due to the scheduling of jobs instead of installations (as Tata Steel currently does) and due to the computerized approach instead of a manual approach. The main advantages are:  Sections have more insight in and a better overview of the maintenance to perform during the year.  The method to generate the YSS is neither based on experience, nor a manual process.  Less additional maintenance jobs appear during the year, because the schedule includes every job. This leads to less rescheduling and more time to prepare a shutdown.  The method takes four objectives into account for each and every job.  The consequences of rescheduling can be analyzed quickly and comprehensively.  The restrictions of the YSS are clear. Hence, the reasoning behind the scheduling itself is clear.
Item Type:Essay (Master)
Tata Steel
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:85 business administration, organizational science
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page