Manufacturing

Introduction

The Internet of Things (IoT) era presents many opportunities, not just pseudo-sentient fridges trying to bring down the internet. Indeed, we can monitor machine activity in real-time to alert operators to immediate problems, track production or the frequency of issues over time, and predict the likelihood of meeting targets in the future. What's more, this approach is extensible; you can start with the isolated case of planning machine production, then start to trace back towards optimising the supply chain for raw materials, ensuring you place orders at the optimal time to avoid stockpiling.

The example below is a brief demonstration of how the data might be collected from real-time monitoring and then utilised to inform decisions on how best to utilise equipment in the short-term to maximise long-term targets.

Real-Time Monitoring

The first stage of simulating production is the collection of real-time data. Fitting machines with sensors provides a product count and the machine status, which can be recorded in the installed programmable logic controller (PLC). This, in turn, can be read and displayed on the shop floor via TV screens (using a PTV , for example, or even just a Raspberry Pi).

In addition to providing immediate feedback to machine operators and supervisors, it's also possible to collect and record machine performance over any period. This allows for reviewing the success of any preventative maintenance programmes or new initiatives.The machines here are fictitious and maybe a little exaggerated in the frequency of their downtime; the code for their activity is.

Machine 1 Product Count:   2468 Offline for:   0 mins 0 secs # Times Offline:   16 Machine 3 Product Count:   2223 Offline for:   0 mins 0 secs # Times Offline:   5
Machine 2 Product Count:   1656 Offline for:   1 mins 17 secs # Times Offline:   9 Machine 4 Product Count:   1283 Offline for:   0 mins 2 secs # Times Offline:   40

Performance Review

Once the issue of real-time monitoring is addressed, attention can turn to performance review. Which machines are causing the most issues? Can a new policy of planned, preventative maintenance (PPM) address these issues or is there a need for more training for operators? Such questions can only be answered by having data, and all the historical data from real-time monitoring can be backed up for analysis.

Production Planning

Whilst it's useful to be able to review machine productivity over time, there are ways in which this data can be put to work. In the example below, we can imagine a fast-moving consumer goods (FMCG) manufacturer that uses machines A through D to make a variety of products, and the company has a sales forecast that it must work towards. Each of the machines can make any arbitrary subset of the product catalogue and the question then becomes; how do they utilise their assets in order to meet the sales forecast in the most efficient way, minimising the chance of failing to meet an order? This is an example of the Job shop scheduling problem and we can build a model to answer this question.

In the panels below, you can make changes to the problem setup and try and solve the problem. You can define which machines make which products, alter the shift patterns of the machines, and define the sales forecast for the coming month. The model will use the production history recorded above in order to determine the output of each machine.

The model outputs will show:

  • The forecasted demand and the machine output to meet the forecast based on their shift patterns and manufacturing schedule
  • A schedule table showing what product each machine should make for each shift
  • A weekly summary of machine utilisation. Utilisation is based only on the shifts that a machine is scheduled to run; the absolute utilisation would be much lower. If the model starts over-producing, the machines will be switched off, indicating that fewer shifts can be used to meet the forecast
  • A graph of solver convergence. The initial, randomised solution will likely be poor at meeting the forecast and will have a high cost. As the solver progresses, the cost will come down and eventually flatline, indicating that no further improvements can be made

Set the shift patterns of the machines
Machine A
Machine B
Machine C
Machine D
Set the product(s) that each machine can produce
Machine A
Machine B
Machine C
Machine D
Set the sales forecast for the next month (1000s of units)
Product Name Week 1 Week 2 Week 3 Week 4
Product 1
64
0
56
96
Product 2
0
196
0
0
Product 3
0
0
0
63
Product 4
0
0
90
62
Product 5
157
0
61
0
Product 6
56
97
61
0
Product 7
0
0
0
182
Product 8
88
144
91
190