Resolving scheduling complexity in a very simple process

Scheduling complexity in premix manufacturer no text

 

Resolving scheduling complexity

By Milos Milenkovic, Postdoctoral Research Fellow at the MIT – Zaragoza Program .

Leer versión en español

 

Some manufacturing processes can be deceptively simple. Take the preparation of premix animal feed. Pour ingredients into a mixer, give everything a good churn, and bag the results. That is basically all there is to it. But it turns out that optimising the lot sizing, scheduling and sequencing of customer orders in this very straightforward set-up is beyond the capabilities of even the most experienced human brain. A leading premix manufacturer turned to ZLC for help.

Premixes are blends of micronutrients such as vitamins and minerals which are added to animal and poultry feed to promote the most efficient conversion of food into saleable product while ensuring the health of the animals, and they come in many different formulations to suit the species and breed, the rearing regime, and the farmer’s commercial objectives. Demand for meat and dairy products (and therefore for animal feeds) is soaring, in India, China and elsewhere with rising incomes and urbanisation, significantly raising the price of feedstocks such as soya. So efficient conversion is an imperative for farmers, while feed manufacturers must do all they can to maximise the use of production capacity while minimising total costs.

That is easier said than done – there are many variables and constraints, as a walk through the production process will make clear.

A customer order will typically generate several production orders – for different formulations, for different bag sizes, and if the quantity required exceeds the batch capacity of the mixer. (Conversely, a single production batch may be serving the requirements of several customer orders). The production planner checks the availability of raw materials, and the due delivery date, and releases production orders to the scheduler.

That triggers two processes. An operator prints out labels for the product bags (which range in capacity from 1kg to 500kg) with details of formulation, batch number, date and so forth. Meanwhile the core production process commences. Ingredients for each batch or lot are weighed and consolidated – weighing ranges from ‘major dosing’ with bulk packages, ‘minor dosing’  of specific amounts from a bulk pack, and ‘precision dosing’ which involves very small and precise amounts of active  ingredients. (Overdosing provides no benefits and may even be harmful, besides raising costs). The consolidated ingredients are loaded into a mixer.

This is where the complexity begins. There is a set of mixing units of different types (horizontal or vertical) and more importantly, capacities (although they all have the same cycle time, of an hour). Production is 24-hour, in three 8-hour shifts. Clearly in order to maximise utilisation lot sizes should be as close as possible to the maximum capacity of the mixer (ie, don’t run a 10kg order on a 500kg capacity machine – in practice, that wouldn’t mix satisfactorily anyway). Equally obviously, the lot size needs to equate to integer numbers of the bags into which it will be packed, so no underweight bags needing to be topped up, and no excess mix that isn’t fulfilling an order.

There are further constraints – some mixers can be worked by one operator; others take two, but there is a limit as to how many workers can be working the mixers at any one time. Additionally, if successive batches are ‘incompatible’, in other words, cross-contamination is strictly prohibited, the mixers need to be cleaned down between batches. This may be a manual operation with brushes and compressed air, or by running the machine while loaded with a cleaning substance – the time penalties are the same but the costs are different. There are also other set-up costs (eg loading) to be considered.

And as with any scheduling and sequencing task, time is of the essence. Some orders have hard deadlines which may not be missed. Other orders have soft deadlines – these can be violated, but at a cost or penalty. Conversely, production earlier than needed is not favoured – this gives rise to storage costs, and the risk of product deterioration while in store.

So it becomes apparent why scheduling and sequencing is not an easy task – indeed, our model of the system contains no fewer than 15 separate sets of constraints, not including cleaning. That complexity can’t be optimised by manual methods, and certainly not in the available timescales. Nor can manual methods resolve lot sizing, sequencing and manpower resource constraints simultaneously.

Our solution was to model the system using Mixed Integer Linear Programming (MILP), and use this to write the software for a PPOptimizer-Premix Production Scheduler – this was developed in C# using Visual Studio programming software, and very much with the needs of the end-user in mind. For simplicity we chose to address the machining issue using post-processing heuristics. The model captures, and the programme processes, minimum and maximum machine capacities, set-up, operational and cleaning costs, and human resource requirements and availability. It thus generates simultaneously lot sizes, schedules, and human resource allocation. Input to the tool is the customer orders as output from the corporate ERP system. The model allows mixing and splitting of production orders, and the user can alter parameters such as the number of active machines, and their capacities, and available workforce data. Run time for the programme was limited to ten minutes.

We tested the optimizer/scheduler against five ‘real world’ instances, ie the schedules that manual methods had derived, ranging from 20-50 orders, for 50,000kg to 100,000kg of product, over horizons of 1-3 days, which is the typical short-term scheduling horizon.

The results were gratifying. In one instance we scheduled a group of 23 orders, which included 45 separate products in 13 ‘families’, totalling 54 tonnes of production. Whereas manual methods resulted in a capacity utilisation of 39%, the optimiser upped this to 76%. Cost comparisons couldn’t be made, due to a lack of detailed cost estimates in the current operational environment, but clearly these have the potential to be very significant.

There is scope for further development: automating the input of data from the ERP, and using the output from the optimizer as input for other planning decisions, such as raw material inventory and finished goods storage capacity.

In truth, this approach isn’t exactly groundbreaking, but there is little evidence in the literature of its use in comparable industrial situations. These must be many of these – paints, specialty chemicals and food production spring to mind, and a much wider field where an apparently simple, even basic, production system has unexpected complexities.

 

For more information contact Milos Milenkovic at [email protected]