Using Artificial Intelligence (AI) to Reduce Scrap and Energy Usage in the Casting Process

Greenhouse gas (GHG) emission reduction requirements are found at the state and country level in most developed countries. For example, in 2020, China announced its goals to strive for carbon dioxide peaking before 2030 and achieve carbon neutrality before 2060.¹ As major industrial emitters face increased scrutiny to reduce emissions, the focus on regulatory compliance will also extend downward to encompass large- and medium-sized enterprises.

By Stephen Cherlet, owner/founder – FarStar S.A.C. Consulting and Dr. Mo Abuali, CEO and Managing Partnert – IoTco

The use of castings is central to a significant portion of the industrial valve industry. Depending on the valve design/type, castings can be found in pressure boundary parts (body, bonnet, body end), wedges, and other key components. Foundries are major energy consumers. Studies² have shown typical steel arc metal casting, typical for valve and pump manufacturing, consumes approximately 20.7 million BTU/ton. Reducing scrap and improving yield, in general, would represent notable energy consumption reduction, and therefore GHG emissions, where renewable energy is not being used in the steel-making process.

The automotive industry can serve as a guide. Predictive quality and energy reduction applications for high-pressure casting have been implemented in the automotive industry (using the Predictronics PDX platform). In this application, the software incorporates process data from the high-pressure casting process, including pressures, temperatures, cycle time, and other variables.

A multivariate machine learning (ML) model is used to detect departures from the normal process conditions and provide insight into the factors contributing to the undesirable quality. Early detection allows the automotive part to be scrapped early in the process, so a re-melt can occur, which is much more cost-effective than scrapping the part at a later stage.

Process adjustments are recommended based on the key contributing factors affecting quality. This allows future runs to have better quality and fewer defects, such as porosity. An overall reduction in yield of 1% up to 3% results in significant cost savings and energy and emission reduction. A typical high-pressure die-casting facility could use approximately 6 million kWh of electricity and 18 million kWh of natural gas per facility per year. This facility would generate 60 million kg of CO2 per year.^2,3 Each 1% reduction in scrap would then result in a reduction in CO2 by 600,000 kg per facility annually.

This foundational work was applied to high-pressure die castings, which do have applications in some segments of the valve industry, typically in aluminum for solenoid valves and various components. This ML/AI approach has a track record in other manufacturing applications, including injection molding, extrusion, and stamping.

For sand castings, a two-pronged approach to process improvement can apply. The typical approach to casting optimization employs tools, like MagmaSoft® for casting and mold design, and the casting process itself for both steel and die casting.

Supplementing the use of such existing tools by leveraging this predictive ML/AI approach for the sand-casting process, for example, is technically feasible. It would require designing the appropriate Internet of Things (IoT) system to monitor the key process parameters and linking these process parameters with quality data. Consideration should be given to extending the monitor parameters to include atmospheric conditions inside the work location and outside the facility, sand humidity, and the other (typical) known variables.

The solution would utilize the process data to detect anomalous conditions at an early stage and recommend process adjustments to avoid sand casting defects (porosity, pinhole, shrinkage, and more) for future runs. Analogous to the high-pressure die-casting example, this technology would assist in reducing scrap, energy usage, and emissions.

A 1% reduction in scrap may not seem like a significant amount, but a broader context exists. The global production of castings, of all materials, in 2020 was 105.5 million metric tons.⁴ This was an increase of 40.8 million tons, or 63%, from production levels in 2000. China, the United States, and India are the “Big 3” producers globally. China’s casting production volume exploded four times from 2000 to 2014 and is still growing. India’s production increased dramatically during the same period but has not reached China’s level and decreased slightly in recent history. The United States has seen a 14% decrease over the 20-year horizon but retains the #2 spot globally. With 49% of current casting production volumes, China is the largest single casting producer overall.

The output for cast steel production, a subset of total casting production, rose in line with the increase in total global output of all casting types. Throughout 2000 through 2014, cast steel production was relatively steady at 8-10% of the total.⁵

The call to action is for metal casters and valve manufacturers to start planning their roadmap for sustainable manufacturing. Foundries should consider how IoT and AI solutions could be used to improve quality, reduce energy, and improve sustainability for sand-casting and other relevant processes. The place to start is with a digital transformation readiness assessment (as-is state), followed by an action plan to attain a future (to-be) state. Savings opportunities could be significant. Buyers of founder products should validate their supply base for programs and activities focused on improvement and sustainability. Long-term supply agreements and potential collaborative efforts, with savings sharing, are an established procurement practice in many industries.