How to use artificial intelligence to assist technology manufacturing enterprises ?

How to use artificial intelligence to assist technology manufacturing enterprises ?

How can artificial intelligence revolutionize supply chain management? Kapoklog Logistics introduced in detail the specific application of AI technology in the supply chain.

 

Through machine learning and operational research optimization, it improved demand forecasting, supply planning, inventory management and order delivery to promote the enterprise supply chain to achieve Kapoklog Logistics, which provided a profound insight into how technological innovation can drive enterprise transformation.

Kapoklog Logistics Logistics shares Lenovo's practical experience in using AI technology to empower the supply chain, hoping that these contents can provide useful reference for everyone and help their enterprises achieve excellent operation in the future.

The sharing of Kapoklog Logistics is divided into five parts. First, introduce the current situation of Lenovo's supply chain. What problems do we face? Why do we need to undergo digital transformation? Taking Lenovo Group, a Chinese electronic technology company, as an example, Lenovo's digital transformation has always focused on building an intelligent supply chain. So, what kind of technological architecture should an intelligent supply chain have? What are the landing scenarios?

 

The supply chain department manager of Kapoklog Logistics will focus on demand forecasting, material allocation and consumption and intelligent scheduling, which are our internal star projects or best practices. Finally, there should be some time. The supply chain department manager of Kapoklog Logistics will also share the big model, agent, which is the most advanced direction of AI. How to deeply integrate AIGC technologies with the real scenarios of the supply chain, solve business pain points, and generate practical value. Finally, based on my very shallow thinking, I would like to make a prospect for the future technological direction of smart supply chain.

 

1. Lenovo Global Supply Chain Overview

Let's get to know Lenovo's global supply chain together. For a manufacturing enterprise, supply chain is an absolutely critical functional department that provides services and guarantees for sales ahead. So what does Lenovo sell? We are the world's largest manufacturer of personal computers, as well as smartphones, tablets, servers, and various smart terminals. With an annual shipment volume of 120 million yuan, we sell to over 180 countries and regions worldwide, covering more than 1 billion users. Such a huge market requires a large supply chain. Therefore, we have more than 30 factories spread across the world, with over 5000 suppliers and more than 2000 core suppliers. In addition to being large, we must also be strong. Therefore, the group invests over 1 billion yuan every year in the digital transformation of the supply chain and has achieved remarkable results.

 

Lenovo has been ranked in the top 10 of Gartner's global supply chain rankings for three consecutive years. We have achieved impressive results in supply chain leadership, ESG, and intelligent manufacturing, and have received full recognition from IDC Mingsheng, including the World Economic Forum McKinsey, and others.

This continuously strengthening supply chain is facing five major challenges at present. In other words, this is also the direction we need to focus on in our digital transformation. For example, in the decoupling of technology between China and the United States, many core components, including some core industrial software, are restricted from purchase in China, and we must rely on self-developed demand fluctuations. Lenovo operates in the consumer electronics industry with small batch, multi variety, and personalized order dates. We receive over 80% of orders annually, with no more than 5 small batch orders.

 

The regulation of ESG countries is becoming increasingly strict, and we will pay attention to some low-carbon technologies. However, with the economic recession lagging behind, the GDP of major economies did not meet expectations, the rise in commodity prices, logistics disruptions, and other geopolitical conflicts and black swan events.

 

So, facing so many challenges, what can Lenovo's supply chain do? It is about embracing various innovative and core technologies, using them to train our internal strength and arm ourselves, so that our supply chain can be prepared for crises before they occur, increasing the resilience, adaptability, and intelligence level of our supply chain.

 

Lenovo's digital transformation of the supply chain has been going on for almost 10 years. Before the pandemic, we referred to it as the 1.0 stage, which focused on consolidating the foundation and visualizing knowledge. From procurement to planning, manufacturing to logistics, and then to service, various functional departments need to collect all data first. Based on the agreed process rules of each department, real-time data visualization and partial decision automation can be achieved.

 

After the epidemic, we entered 2.0, which has just been launched. The focus is on interconnectivity, collaborative intelligence, connecting various data silos, establishing connections between functional departments' own small systems, and achieving overall win-win results. Therefore, we must redesign processes, redefine rules, and unify them. Data needs to be standardized, not only to have data, but also to turn it into high-quality data, achieving fully intelligent and comprehensive solutions. The biggest feature here is to promote proactive decision-making driven by data analysis, helping our decision-makers and planners to be proactive and have foresight before crises arise.

 

2. Lenovo Functional Supply Chain

So we just mentioned that we should arm ourselves with various cutting-edge technologies. Here I list the eight innovative technologies most concerned by Lenovo's supply chain. Automation is the robot here, but we can still classify them, such as the Internet, blockchain, etc., to provide data or guarantee reliability. Digital twins provide an environment where our various simulation algorithms can continuously evolve. Advanced analytical artificial intelligence focuses more on the models and algorithms themselves, and how to extract insights from data to guide the future.

Taking advanced data analysis as an example, we generally divide it into three levels. Firstly, the descriptive analysis tells us what is happening now. Secondly, the predictive analysis tells us what will be discovered and what will happen in the future by analyzing the patterns and patterns in historical data.

 

The third stage is called decision analysis, which not only tells us what the future trends are, but also tells decision-makers what they should do. Artificial intelligence has a history of more than 60 years, and most people may think of computer vision, speech recognition, natural language processing, machine learning, and so on. Those technologies are impressive, but one thing that is often criticized is that it is an unexplainable black box model.

Fortunately, the accuracy and performance of these technologies have now reached a certain level, and in some focused fields, they can reach or even surpass human levels. So why not embed it as a plugin into the workflow of the supply chain.

 

For example, automatic optical inspection on the production line can identify some defects in products and installation defects, and its accuracy far exceeds that of human eyes. Will so many technologies make people dizzy and at a loss? This problem does not exist at Lenovo. Because we are all driven by scenarios and requirements for each project, and then we look for suitable technologies to finalize it.

 

Here is a simple example, which is Lenovo doing all digital transformation projects. In scenario based solutions, the various core technologies mentioned earlier will settle into the end-to-end supply chain control tower. The surrounding 12345 is a typical scenario. For example, in the first scenario, demand forecasting used time series analysis, machine deep learning, and AIGC.

 

The second intelligent procurement and supplier highly collaborative hierarchical management, in terms of researching suppliers, we have accumulated a large amount of procurement data to determine whether suppliers are reliable and whether delivery can be on time. At the same time, we also use natural language analysis and data mining techniques to analyze the 360 degree profile of the supply chain from publicly available data in the periphery.

The intelligent management of the third customer order, we have so many orders with different product quantities, delivery times, supply and customer levels. We can use optimization technology and operational optimization technology to intelligently decide which orders to execute first, which orders need to wait, and even ensure the smooth implementation of specific orders.

 

The last intelligent logistics, including delivery, warehouse location selection, logistics method selection, and which station to deliver the last mile first and which station to deliver last, are all some operational optimization technologies.

 

Everyone can focus on the red banner and the yellow text covering the three major processes, from demand to supply in the planning stage before the first order arrives. The second order has reached the delivery stage, from order to cash. The third aspect that revolves around the entire product is the optimization of its lifecycle, the three major processes, and the eight major technologies mentioned earlier.

 

We summarize it into two core technology categories, one is called predictive technology represented by machine learning, which aims to solve how to reduce uncertainty, or in other words, how to use data and algorithms to make some previously very uncertain scenarios slightly controllable. Let me give you an extreme example. When we throw a coin, without any data, it's 1/2. However, if you know the angles of the throw and even the different textures of the characters on the front and back, you can still make more accurate predictions through data analysis.

 

Another major category is decision-making techniques represented by operations research optimization, which is used to solve multi-objective balanced supply chains, transportation, even asset allocation, and energy resource allocation, all of which involve different conflicting multi-objective problems. How to achieve optimal balance? Traditional operations research can solve such problems.

 

We are in the manufacturing industry, especially in discrete manufacturing. We have a concept called 'bill of materials', which can actually be managed in a layered manner. The more we are at the top of the first layer, the more critical it is. By following the diagram, we will definitely be able to find the best core technology for a certain item, which is strongly connected to the pain points of a particular scenario we are concerned about.

In the field of discrete manufacturing, we have the concept of a 'bill of materials'. The components can be managed in layers, with the first layer closer to the top being more critical. Through this hierarchical management, we can identify the most critical materials in a specific project based on established clues. This core technology is closely linked to the pain points of the specific scenarios we focus on, and can effectively help solve practical problems.

 

3. Case study of supply chain intelligent brain

There is a concept here called the supply chain control tower, which is the supply chain brain we mentioned later. Here, I will give one or two examples to illustrate the concept of furniture. The first technology is prediction technology, which can be used in many scenarios, such as demand forecasting, sales forecasting, brewing forecasting, production capacity forecasting.

 

The purpose of prediction is to solve uncertainty. What are the characteristics of our Lenovo prediction? It's not just about predicting and estimating with numbers. Our characteristic keyword is called mixing, which is specifically reflected in the mixed use of AI at multiple levels. For example, you can use a combination of algorithms, statistical methods, machine learning, deep learning, and a mixture of large and small models to run in the cloud. Other GPTs can also be tested at the edge. For example, on the production line, we need to detect the uncertainty of those signals, solve it at the edge, and then support the combination with data.

 

In addition, there is multi-level integration. We need to make sales forecasts. We can have various regions, each with different countries. Should we divide them after the top-level forecast is completed, or aggregate each local forecast? In fact, there is no one size fits all approach. We need to make dynamic judgments based on the distribution, quality, and format of the data. Returning to the combination of knowledge and data we just talked about, let me give an example. For example, when we sell PCs at Lenovo, some of them are sold directly to customers, but most of them are sold through channels. As for how much the channels sell to end customers, we don't know, but we really want to know.

 

Because the trend of sell out and the inventory of the channels determine how much we ship, how can we predict sell out? Of course, we can ask the channels to collect how much they sell every year, every month, and every quarter, and measure it with numbers. However, the actual effect is very poor. Later, we found that we lacked some key information. This is what the sales told us. You scientists should go and understand the opening of rebates in various channels. If he can sell 80 units and get a 5% rebate, he will sell 81 units, 85 units 82 units, but he definitely won't sell 79 units. He must reach 80 units. Once we grasp this pattern, the accuracy of our predictions can be greatly improved.

 

For example, when we used to make predictions for the service supply chain, it was called 'how many spare parts should maintenance stations prepare for emergencies'. We encountered a problem in Southeast Asia at that time, such as India's inaccurate prediction of its display module from May to August. Later, the local business told us to study the weather carefully. May to August is India's monsoon season with high humidity, and the failure rate of devices with built-in circuits is quite high. How high is it? He doesn't know, but through his prompt, we know how to extract temperature information, rainfall information, humidity information, and fuse these data together. During the rainy season in Southeast Asia, our prediction accuracy has greatly improved.

 

Let's take a look at the second application area related to optimization. I have prepared some videos, and I will talk a little less. Everyone should understand after watching the videos. Before delving deeper, let's briefly understand the concept of "optimization". From the perspective of operations research, optimization mainly involves three major elements. Firstly, in the supply chain scenario, there are many objectives that need to be optimized. For example, the delivery rate, cost, transportation route, waiting time of orders, including the time for changing lines between different products on the production line, etc., are all business goals that we need to focus on and optimize. This is the first element of optimization.

 

The second element is the decision variable. For example, when an order is placed or the requirements are specified, we need to purchase parts for assembly. At this point, we are faced with a choice between purchasing from supplier A or supplier B, assuming that the components from these two suppliers can be replaced with each other. Not only that, the time and quantity of procurement from supplier A also need to be carefully considered and decisions made. For example, in the scheduling process, whether the work order is arranged on production line A or production line B, different choices will have different impacts on production efficiency, costs, and so on.

 

The third element is the constraint condition. We all know that resources for doing anything are not endless and must be carried out under certain constraints. Taking the example of the previous work order, it is possible that a certain product can only be produced on the ABC production line due to process, equipment, and other reasons, and cannot be processed on the D production line.

 

This is a typical constraint condition.

For example, in the case of retail stores, they can only pick up goods from local distribution centers and cannot allocate goods from neighboring provinces, which is a constraint. The solution obtained under these constraints is the feasible solution. However, not all feasible solutions can satisfy people. After we have a clear goal, we need to find the optimal solution. Even if there are multiple optimal solutions, we still need to find suitable points on the so-called Pareto front.

Let me give you an example of intelligent material allocation. During the epidemic, supplies were in short supply, and CPUs from companies like Intel or AMD were in high demand. Various regions and customer groups requested these popular CPUs from the headquarters' supply chain department.

 

So, who should these precious CPUs be allocated to?

When there is no intelligent algorithm, the usual approach is to use equal distribution, dividing everyone's total demand by the total supply. Assuming that each person can receive 80% of the demand, it appears that everyone is quite satisfied on the surface. However, the boss doesn't think so. Because the profits and revenues of products vary in different regions, and there are also differences in customer tolerance.

 

So, we need to integrate financial goals, fairness goals, and other objectives to conduct multi-objective optimization, in order to allocate these CPUs or scarce materials such as display modules reasonably.