Classified incorrectly, dangerous goods could endanger people’s health and safety. But few experts possess the knowledge to constantly update a large enterprise’s portfolio. Using the power of machine learning, an intelligent supply chain management solution from SAP automates the classification and declaration of dangerous goods.
Thermometers contain mercury. Fertilizers are flammable. The airbags that might save lives in a car crash can explode when not built into a car. Many goods indispensable to our everyday life are actually labeled as dangerous, which makes their storage and transportation a delicate matter.
“There is little awareness of how we are surrounded by dangerous goods that are being used for benign purposes,” says SAP’s Volker Loch, “which shows that safety regulations are doing what they’re supposed to.”
Indeed, chemical substances are subject to strict and complex regulations with regards to product safety and the environment. What remains crucial is the correct classification and declaration of dangerous goods and the ability to move them safely and compliantly through the supply chain. The transport of such goods is governed by a complex network of national and international regulations that have broad applicability across many industries.
“Incidents that stem from dangerous goods lacking classification can reach from mere nuisances to full-blown catastrophe,” says Loch. Incorrectly declared and therefore incorrectly stored, they can start to leak and react with other substances close by, even resulting in combustion in the worst case.
“Planes have been forced to make emergency landings because of substances being distributed via air conditioning when a container broke,” Loch explains. “The companies shipping those freights needed to pay thousands of Euros.”
Chemical Industries Need Experts
Knowledge of how to classify dangerous goods is a rare skill, and it’s getting even rarer as the job becomes more difficult. Big companies have thousands of products in their portfolios. The classification procedure is repeated every time something is changed about the product.
“Experts are working to capacity and they are basically doing the same thing over and over again,” says Julian Stoettinger, a data scientist with the Deep Learning Center of Excellence at SAP. It’s not easy to attract employees to this kind of task, especially when it comes to the younger generations. “It’s become difficult to find experts in dangerous goods because junior employees are rarely interested in this line of work. They prefer not to work with dangerous goods.”
Transferring their expert knowledge on dangerous goods and how to classify them into a machine learning model is very attractive to a number of SAP customers.
How Machine Learning Complements SAP EHS Regulatory Content
SAP EHS Regulatory Content packages have provided a rule-based system to support experts in classifying dangerous goods. However, the complexity of international regulations, industry-specific exceptions, and the vast number of cases do not allow defined rules for all known cases. This is where machine learning and deep learning come in.
The Deep Learning Center of Excellence is part of SAP Leonardo Machine Learning. The team at the center works on a variety of machine learning use cases, collaborating with various product teams across SAP to take existing solutions to the next level with machine learning. Together with the development team for SAP EHS Regulatory Content, they developed two enhancements to SAP S/4HANA for product compliance:
- The chemical substance Group classifier correctly assigns substances to substance groups in order to secure compliance with regulations. The Deep Learning Center of Excellence created a deep learning model for the most important substance groups, based on existing substance group assignments in SAP EHS Regulatory Content. More than 300 substance groups can be proposed given a substance name.
- The dangerous goods classifier differentiates between dangerous and non-dangerous goods and detects all relevant hazards for dangerous goods, building the foundation for the final classification by experts and increasing product compliance. It was built using the data originating from existing dangerous goods classifications for products in the SAP EHS Regulatory Documentation OnDemand service, and also detects different classifications depending on the regulation, for example for specific regions or means of transport.
SAP customer Merck aims for automatized classification of dangerous goods as a co-innovation project with SAP.
“Our product line contains roughly 1 million chemical goods,” explains Andreas Ehlers, head of Digital Transformation, Hazard Communication and Chemical Regulations at Merck Life Science. “Currently, we are using a manual approach for the classification, which means a high effort due to the large number of products. The implementation of a rule-based approach is under evaluation. However, this means a high effort in configuration for us. We would like to have better automation, that is what makes machine learning interesting for us.”
But not only new or modified products are supposed to be checked by the dangerous goods classifier. “We plan on using the solution for double-checking all our human-appointed product classifications,” says Ehlers.
Currently, experts from SAP are extending the machine learning model with data from 300,000 Merck Life Science products. As a result, Merck will receive a software calibrated to their product portfolio and other customers will profit from the model’s new capabilities as well.
“Classification isn’t meant to be fully automatized; a machine cannot and must not take over this kind of responsibility,” says Ehlers. “But receiving suggestions for classification from an intelligent assistant will save us lots of time. The quality of the classifications will increase through automated controls.”
Machine Learning Enabling Chemical Industries
The dangerous goods classifier will support experts in their daily tasks, allowing them more time to focus on the complicated cases. Classifying dangerous goods with 99.8 percent accuracy, it works as a second control instance for classification.
“The solution is basically an enhancement of human judgment and can correct human mistakes made in highly complex situations,” says SAP’s Julian Stoettinger. “Applying machine learning models not only allows significantly improved compliance for chemical products, it also minimizes risk for our health and environment.”
By integrating machine learning, SAP S/4HANA for product compliance not only enables chemicals industries to create better, safer products, it also allows employees to focus on tasks that require creativity and the advanced problem-solving skills still unique to humans.