Littlefield Round 1 Analysis

Littlefield Round 1 Analysis

Assumptions:

• Flow rate through system is at steady state. ([pic 1]
• Used linear regression to forecast demand.

Using the first 50 days of data, we used linear regression to forecast demand. The average demand per day was 2.64 jobs with the forecasted demand hitting 9 jobs by day 135. Station 1 also had a high average utilization of 0.51 and an average queue rate of 18.9 kits. Thus, we decided to immediately buy an extra machine for station 1 once the simulation started in anticipation of higher demand while monitoring station 3 as it also had a relatively high average utilization of 0.388.[pic 2]

[pic 3]

At day 54, we noticed a spike in queue rates in stations 3 and 2 resulting in us purchasing additional machines in both stations while changing the priority of station 2 to step 4 as step 4 required more time than step 2. The purchases managed to bring down the queue rates and bring us to the top of the simulation. However, we did not continuously monitor the demand and change the linear regression model accordingly. Based on the initial forecast from the first 50 days of data, we assumed that we could handle the demand until day 150. However, our assumption was far off and the demand spiked at day 120. We immediately purchased a third machine for station 1 but we were too late and the queues at this station caused us to lose a lot of revenue. The queues eventually cleared but we could not recover the lost revenue and our position dropped significantly. We also assumed the demand would flatten out after day 150 but it continued to increase causing us to lose more revenue and create higher queues at our stations. This was our worst period.

What we should have done

After considering our mistakes, we realized we should have calculated the average capacity of each machine for each station after the first 50 days. From there, we should have continuously monitored demand every 10 days and changed the linear regression accordingly to forecast demand. The first 50 days of data was not an accurate representation of demand. We could then estimate which station would be the constraint and consequently add machines when the demand increased.

Based on the demand pattern, we should have bought the third machine for Station 1 at day 110 instead of waiting until day 120 when the demand had already increased rapidly. We may have needed to add a fourth machine due to increased queues during days 111 – 170. However, we would have had to weigh the cost of the machine versus the revenue gained with buying the fourth machine.