Development of a Pull Production Control Method in the Metallurgical Industry

Read this article. It covers how understanding a supply chain can improve internal production processes. Pay particular attention to the section that outlines push versus pull. Can you compare and contrast each system?

Applying the Model for Production Control to the Metallurgical Industry through Simulation

In order to make the decision and to increase the acceptance for future implementations in ETO manufacturers, an agent-based simulation was carried out in order to show the potential of this approach. The goal of this simulation is the comparison of today's common production control logic, the push control logic, with potential pull control logic.

Process Recording and Data Collection

As explained the methodology the first step was to represent with a process mapping the material flow of the ETO metallurgical producer and associating the data to the specific processes and product families. The important representatives of the individual product families were depicted in a process mapping tool and then brought into the simulation model with their respective production routes. For each product family, the needs of five years were taken to design the base scenario. The agents that go through the simulation were the individual manufacturing positions or units with their parameters such as tons, volume, processing time per machine group, order release date, completion date, the transport times between machine groups, etc. In addition, these positions run through the various machine groups along its production. Each machine group includes a number of machines that manufactures the same products. The machine groups depend on their shift model, usually 15 or 21 shifts as explained before.

Model Formulation

Later, the logical differences between the models are to be described. Both models have the same assumptions as well as the same framework conditions:

  • Time horizon: The tasks are assigned according to their temporal relevance at different planning levels. According to the St. Gallen management model, the strategic planning level has a planning horizon of several years. Based on it, five years were chosen for the simulation of the two models.
  • Production mix: Seven different production families with 12 different production routes were considered.
  • Finite capacity for all production steps. Production capacity as the sum of the capacities of the machines within a machine group.
  • Processing times depend on the product family, on the variant of product family and on the weight of the production unit.
  • Same transport times between buffers and machine independently of the product family.
  • Infinite transport capacity between machines along the production process.
  • Infinite stock capacity before, along, and at the end of the production process.
  • Raw and operating materials are always available for the production process.
  • Assumed that in 3 weeks a batch for a steel type can be produced.
  • Quality problems or rework lead times are not considered.
  • Personal planning not considered.
  • Depending on the machine group, 15 or 21 shifts per week.
  • Production units are the agents that flow along the production process.
  • If the load of a bottleneck is too high, the units wait to get a release date. If there are a certain number of units waiting for it, the manufacturer loses the demand that influencing the bottleneck until the number of units falls below the limit.
Table 1 describes the main differences between both models. The push control starts the production in the steel mill, as far as the orders are in the order intake, since it follows the principle that the machines must not remain still. This causes the production orders to start too early or too late, usually too early, and as a consequence, the WIP increases. In addition, the delivery date determination is not determined on the basis of the bottleneck resources per product family. This, together with a local optimization in the individual production steps, leads to situations with high stock, long lead times and poor on-time delivery.

Table 1. Differences between push and pull production control models for a metallurgical manufacturer.

No. Difference Push-Control Pull-Control Applying TOC/DBR
1 Order release Order release for improving capacity utilization in the first production steps Regulated order release control based on the system load
2 Determination of delivery dates Same order release for units of the same product family Adjusted based on the system load
3 Sequence planning First in First out (FIFO) Global Priority Rule: Priority determined based on time consumption on the delivery date

On the other hand, in the pull control, order release into production is based on the load of the bottleneck resources. For each product family, there are certain typical bottleneck resources to analyze (e.g., indirect stock and direct stock before a machine group converted to processing hours). Indirect stock of a machine group is the stock that has to be processed by this machine group, which is already in production and that is not ready to initiate production in this machine group. By knowing the total amount of processing hours needed to process the direct stock and the indirect stock, the model can determine which one of the machine groups is a bottleneck resource. Therefore, delivery dates are determined by dynamic lead times, which are determined on the basis of the system load. The third difference is the introduction of a global priority rule, which forms the basis for sequencing before each production step. The priority of a position (%, as a percentage) is determined on the basis of the time consumption at the delivery date, the higher the priority the sooner the position has to be processed in that specific production step. This means that at every time period priorities are recalculated to create a list of next production units to be processed in each machine group to optimize the system globally by processing the critical units according to promised delivery date at first.

The two models have moving bottlenecks depending on the product mix that are in a particular moment in production process. The pull-control model can anticipate the bottlenecks by knowing the processing time needed in the current bottleneck resources. Based on this processing time a reliable delivery date is given.
The previously described characteristics are shown in Table 1.

Model Programming

The simplified production flow simulated can be seen in Figure 4 below:

Figure 4. Simplified production flow of the simulation model: 6 main steps.

The models were created using delay times as lead times for production processes, transport, waiting times, etc. The AnyLogic software allows providing a processing time or delaying time depending on the agent entering in a machine group. When an agent or production unit enters a machine, the processing time that depends on the agent is activated. After this time the unit goes out of the machine and waits to be transported to the next processing step.
Moreover, the agents within the simulation are the production units with their parameters associated to them as listed below:
  • Product family (number);
  • Variant within the product family (number);
  • Weight (tons);
  • Processing lead time per production step (days);
  • Days until security buffer (days);
  • Transport lead times between all combinations of production steps (days).
Based on the flow of a production unit along its production process, the following parameters are calculated dynamically within the agents:
  • Waiting time until order release is given (days);
  • Promised production lead time (days);
  • Production lead time (days);
  • Transport lead time (days);
  • Waiting times along the production process (days);
  • Days before or after the promised production lead time (days).
Demand is created within the model with gamma distribution and is equal for both models. Its value is quantified according to the rate of incoming units or positions of a certain product family and variant and it is recalculated every 90 days, assuming that a certain pattern remains for 90 days.

The key performance indicators (KPIs) for the simulation model are:

  • Demand: production units ordered (units). The value is written in an Excel file in periods of 90 days.
  • Production units started: production units released (units). The value is written in an Excel file in periods of 90 days.
  • OTD (on-time-delivery): percentage of units produced before the promised delivery date for logistics (%). The value is written in an Excel file in periods of 90 days.
  • Production throughput: cumulated production (tons).
  • WIP: quantity of units in production process (units).
  • Stock before and after production process: quantity of units before and after the production process (units).
The KPIs per production family are: demand (units), OTD (%), production lead time since order release (days), lead times for technical processing, transport and waiting times (days), percentage of units that did not reach the security buffer (%), and the weight break-down of the units (tons and %).

The KPIs per machine group are: capacity utilization (%), planned load (weeks), priority of a unit before a machine group (%), and direct stock (weeks).

Model Testing and Simulating

The models can be initiated with WIP production units or without them depending on the adjustable parameters. Moreover, this and other adjustable parameters are shown in Table 2:

Table 2. Adjustable parameters.

No. Adjustable Parameter Description Unit
1 Demand Expected value and deviation based on gamma distribution Units per week
2 Maximal load per machine group Production system load Days
3 Production lead time Time for production for the next orders Days
4 Security buffer Time for buffer for the next orders Days
5 Quantity of machines Number of machines per machine group Machines
6 Shift model Determines the planned production time Shifts per week
7 Performance factor Considers availability and performance losses %
8 WIP at time = 0 Quantity of units in production process at day 0 Yes/No

Moreover, the values for the described parameters are to be introduced in the cockpit of the simulation model within the AnyLogic software, with the exception of the shift model that should be introduced in other screen within the model, as depicted in Figure 5:

Figure 5. Cockpit of the simulation study for the push-control model.

In addition, Figure 6 shows the testing process performed within the simulation software:

Figure 6. Testing the model with the extreme-value test.