Risk-Resilient Plans Using Machine Learning
and AI Technology
Even though AI technology has been with us for decades, only now can we apply them to business functions, thanks to faster processors and availability of memory for big data. Using ML, a system can be trained to predict potential issues and make recommendations to prevent unwanted outcome or to enhance the desired results. There are plenty of examples where this can be done: from operations to sales forecasting and inventory optimization. ML can also be used for diagnostic purposes to identify problems and then figure out causes. For example, an equipment is not operating at the assumed utilization rate even though it is a bottleneck resource. Or the cost of making a certain product is higher than assumed. A cause for the latter may be the use of alternate resources or substitute and more expensive part numbers, used frequently. Perhaps because the designated original part number is not available.
At Adexa, our strategy is to use not just planning data but also transaction data (e.g. ERP and MES) as well as exogenous data such as streaming data for weather, web sites and other events. The data is examined by our ML algorithms in 3 ways depending on the temporal value of data. The figure below shows the approach at a high level.
Self-Correcting uses Past data in order to find trends that are changing the actual model of the supply chain. This approach helps to keep an accurate DIGITAL MIRROR©, or the digital model, of the supply chain. Examples are equipment availability, supplier lead-times during different months of the year, number of rush orders from a subset of customers, increase in carbon footprint etc. The data comes from systems of record such as ERP, MES and PLM.
Self-Improving uses past and future (or planning) data to identify frequency of issues and how to improve supply chain policies. Examples are the right level of safety stock for each product at the right time or having more focus on cost of certain products and changing supplier and customer behavior with incentives to get better service and faster response from them.
Self-Optimizing means the ability of algorithms themselves to self-improve their performance and efficiency. In other words, we do not need to apply the same heuristic and/or optimization method every time. But over time the algorithm can perform better because it can deploy some of the specific properties of the data in that environment. An example of this is use of “markers” in “search” techniques. The markers leave a message behind for the next search to indicate if it is a good idea to follow a path amongst so many other choices. Consider playing a chess game where you have so many moves to make. Given the same conditions, others have made better moves that you can follow and or avoid certain moves based on other players’ experiences. Self-optimizing learning works the same way avoiding making moves in the wrong direction. Obviously by doing so search techniques can be performed much faster and much more efficiently. Just like chess, you have to react in more or less in real time!
Use of Attributes as AI Expert Systems
Adexa attributes and Attribute Based Planning is used to learn business rules from the users and planners. The attributes form Boolean expressions that define such business rules. The latter is then used as a constraint in the supply chain when plans are generated. Examples are use of certain parts for certain customers, use of qualified suppliers for certain orders, keeping cost or carbon footprint at a certain level and pegging certain features of the intermediate product to certain orders. Attribute-Based Planning (ABP) makes the system very flexible and allows it to constantly adapt to the changing environment of the supply chain without having to make code changes.
Temporal Value of Data
We can categorize data by its temporal value: Past, Present or Future. Past data may come from transaction systems and can be used for diagnostic and predictive purposes as well as causal analysis. If we have an accurate and detailed model of the supply chain (A Digital Mirror©), then every time a plan is made, we can observe what potential issues we have and how frequently and why it is expected to happen. Over time as we make supply chain plans, certain trends emerge. Such trends are picked up by the system and used to avoid risks and change policies as described above. Present, or real-time, data may be treated differently depending on how fast a response is needed. Examples are an equipment breakdown, large order received abruptly, or a supplier delivery does not arrive in the morning as expected. This data would be used later on if a pattern emerges, however at the time that it happens, an intelligent system, machine or man, can react and prescribe what to do. For example, re-plan or use substitute part# or call the supplier amongst others. In the section below Adexa Genies© are deployed to specialize in such tasks.
Scalable and Flexible Approach to Deploy AI and ML
In order to deploy the above technologies in our application, Adexa uses distributed and independent agents (processes) to perform specific tasks and to learn from their environment as well as communicate with relevant stake holders. These Agents are called Adexa Genie©. In a typical supply chain, many Genies can be deployed to perform their designated tasks. Genies get better over time and can perform more sophisticated functions as they learn. Click to learn more about Adexa Independent and Distributed Agents, or Genies, in more detail.
Beyond the Enterprise
A white paper on how Adexa solutions make suppliers and contract manufacturers more responsive, and accountable.
Evolution and Supply Chain Planning 2015
You don’t have to be an evolution scientist or even a believer to know that being big does not necessarily imply strength and endurance. I t does not even guarantee survival.
One way to define intelligence is the ability to predict the outcome of different circumstances and/or construct a scenario to get the desired outcome. If all the situations are pre-defined and outcomes are pre-established then this would require less or no intelligence.
Supply Chain Intelligent Machine
The word “intelligent” is used commonly to indicate how well a software package was designed to respond to different conditions. The more conditions it is programmed to handle, then the more, so is claimed, intelligent is the software package.
Supply Chain 2020
If the next 10 years bears even the same kind of growth and innovation that we have experienced in the past decade, then we are going to be witnessing an incredible era in transformation of supply chains and how the age of real-time connectivity will create virtually instant delivery of what we desire.
Industry Directions Solution Profile
Although the economic recovery is well underway, a pragmatic approach needs to be taken when analyzing supply chain solutions. Proof of implementation scenarios need to document the time they’ll take to achieve objectives, the return on investment period and their overall impact on manufacturing assets.
Self-improving supply chains
1st generation SCP systems were factory planning systems or APS, introduced in mid-80’s followed by supply chain planning as the next generation. More recently the 3rd generation introduced sales and operations in an integrated environment which was later improved to the 4th generation by adding financials into the equation forming the 4th generation of SCP systems.
From Predictive to Prescriptive Analytics and Beyond…
Predictive analytics looks at the patterns of the past and predicts the likelihood of an event in the future. In addition, pattern recognition can be used with the use of AI techniques and topology to find relationship between different variables and how one can cause the other.