- What is machine learning and how does it relate to AI?
- Where is machine learning used?
- How does machine learning relate to IoT?
- How can machine learning be used in warehouse management?
- Will machine learning replace the role of operators and managers?
- What are the benefits of machine learning for warehouse management?
AI, machine learning (ML), and IoT are going to have a major impact on supply chain and logistics over the next ten years. This post separates fact from fiction and answers a few basic questions about these technologies. It also describes how ML can be used in warehouse management.
1. What is machine learning and how does it relate to AI?
Artificial Intelligence (AI) is a broad topic that is receiving tremendous amounts of hype. This is leading to abuse and misuse of the term in the service of marketing. For example, a recent study found that forty percent of European AI startups don’t actually use AI.
Putting aside the differences between marketing and reality, an AI expert from Carnegie Mellon University notes that “AI is a moving target.” Fifty years ago, a chess-playing computer was cutting-edge AI.
Today, many applications of AI (speech recognition, Netflix video recommendations, etc.) use machine learning. Machine learning is a process in which learning algorithms are applied to large sets of data to create a predictive model to improve planning and decision making. The model adapts and improves over time as new data is received. Popular ML techniques include deep learning and deep neural networks.
The important point is that systems using ML do not require explicit programming to perform an operation. An example of explicit programming is software code that follows specific steps to determine what move to make on a chess board, or how to calculate the time it takes to perform a given activity. ML systems use algorithms that analyze data to come up with their own answers, and adapt and improve with experience.
2. Where is ML used?
Machine learning is used for things like facial recognition or speech recognition, and it is also used in more mundane tasks like offering video suggestions on Netflix, or shopping suggestions on Amazon. In the supply chain space, ML is being used in multi-site inventory planning, and robotics systems use machine learning to master complex tasks like navigating a warehouse. In the realm of warehouse management, it could apply to a range of functions, including workforce planning and slotting.
3. How does machine learning relate to IoT?
Machine learning requires vast amounts of fine-grained data and tremendous amounts of computing power to process the data. The data required typically does not come from enterprise computer systems like an ERP or WMS that capture and store general transaction data.
Because of its reliance on large streams of fine-grained data, ML is often associated with IoT (the internet of things). IoT refers to interconnected machines (conveyors or sorters with sensors, etc.) and devices (printers, mobile computers, etc.) that collect and share massive amounts of real-time data. ML is an increasingly valuable tool in discerning patterns and finding meaningful information using the masses of data generated by IoT devices.
4. How can ML be used in warehouse management?
In warehouse management, machine learning can be used as an alternative to traditional planning and optimization tools that rely on explicit process modeling and engineering.
One example of the traditional approach is labor management systems that are based on engineered labor standards. An ELS-based system is explicitly programmed to calculate expected work completion times (for a given task or group of tasks) using a pre-defined model and a defined number of variables. ELS requires a significant upfront investment in engineering and measurement to come up with the values used in the model (what is the average time to make a pick, travel x distance, set up a cart…).
By contrast, ML can be used to predict how long it will take to complete a given task by analyzing streams of fine-grained operational data, without the need for the upfront engineering and measurement to determine average values. The data can come from a number of sources, including work execution systems, mobile devices, and automation systems (or WCS). The model will account for indirect influences on results, and it will detect and adapt to changes in the system.
One of the chief limitations of an engineered model is that it needs to be updated to account for any relevant process, layout or other changes that may affect the calculations. Likewise, it cannot readily account for all of the indirect variables that may affect results. Finally, the more complex and comprehensive the engineered model, the longer it takes to process the data and provide an output.
5. Will AI replace the role of operators and managers?
One of the leading misperceptions about AI in general is that it will replace management decision-making. On the contrary, ML tools will provide input to management planning, rather than automating those functions. In the labor management example, above, ML provides an alternative method for predicting labor requirements without the detailed process modeling and engineering required in a system using ELS. Even with ML, there will be a need for management input to determine and manage target rates.
6. What are the benefits of machine learning?
The key advantages of machine learning compared to traditional engineering approaches are:
- Better accuracy. ML systems account for individual differences and indirect factors that affect results, leading to more accurate and useful predictions.
- Faster and lower-cost to implement. ML systems eliminate the need for a labor-intensive modeling and measuring process. For example, ELS systems require a combination of on-site observations, software calculations, benchmarking, and validations.
- Flexibility and adaptability. ML systems learn and adapt to changes in an operation. Engineered systems require constant manual calibration and updates.