Optimizing the manufacturing of Cold Rolled Steel products
The Challenge
Our client, a major manufacturer of cold rolled steel products, had a problem. The raw material used by them cost approximately 85% of the cost of manufacturing the end products and varied in grade and size. This high capital investment in raw materials and consumables results in a low margins. External macroeconomic factors impact the costs of acquiring the raw material and consumables and hence it is important to ensure that forecasts are accurate and machine down time is low.
Hypothesis
- Historical data on orders provides implicit signals to forecast future demand
- Price of consumables and raw material can be predicted using macro economic factors and industry reports
- IoT data from the manufacturing process can be leveraged to estimate health of the production line and help manage downtime
Execution
- Extract historical orders and fulfilment data from SAP
- Scrape macro-economic data, industry reports and news
- Neural architectures using RNN/LSTM and Attention based models were developed to forecast orders and predict future pricing of consumables and raw material and machine downtime
- Optimization of raw material purchase to maximise profit
Outcomes
- Intermediate results show positive outcomes from a forecasting of orders and predicting machine downtime.
- IoT data collected from manufacturing process is predictive of downtime.
- Neural models are improving forecasting of orders received.
Project Highlights
75%
more accurate predictions for order volume and downtime.