The complexity and interconnectivity of today’s manufacturing and purchasing and supply management (PSM) systems are paving the way for new technological advancements in the manufacturing and purchasing and supply sectors. Recent developments in artificial intelligence (AI) and the extensive amount of generated manufacturing data, known as big data, are allowing the integration of new kind of analytics tools in the supply chain, which are optimizing the way goods are produced. The focus of this paper is the application of such AI systems in the manufacturing and purchasing and supply management processes in factories, leading to concepts like smart factory and smart manufacturing, and the restructuring and digitalization on the production floor, dominated till to now by the human workforce.
Advances of applied artificial intelligence associated to the smart manufacturing concept
Data analytics together with AI technology is a combination, which can improve efficiency, product quality and
can increase safety in the production process. The manufacturing sector is thus one of the main driving factors pushing the development of applied AI technology. In these section six areas, which are undergoing a rapid digitalisation and optimization and highly influenced from AI algorithms, are considered.
1- Quality assurance inspections
Using advanced imaging comparison technology, organizations such as BMW and Canon have utilized
artificial intelligence solutions, which can distinguish inconsistencies and anomalies from product design standards and catch defects that would not be undetectable to the human eye. Bosch was able to embed AI into their manufacturing process, which improved the cycle time and automated defect type detection, visual inspection success rate and the quality inspection performance. They also claim to have reduced their CO2 emissions from their manufacturing plants by more than 10% over the past two years and the testing time of their products was reduced by 45% saving them $1.3 million. With the new AI systems in place the reported escape rate of components with defects is at 0% and a false alarm rate of less than 0.5%, whereas with the old systems with mainly human inspectors, the error rate was ranging from 20% to 30%, due to various reasons like optical illusions and imprecisions of eyesight. With such systems the quality stays constant, which benefits business by increasing their customer satisfaction, while the production time and costs involved for a product or a service decrease. Several approaches in machine learning are available for such methods. Supervised learning can be used to differentiate between certain characteristics for products that have only a limited number of features. With sufficient data available, it is therefore possible to perform a classification task and thus find quality defects more quickly. These classifications can be optimized by neural networks and trained to near perfection. The problem with classification and supervised learning arises when states occur that cannot be clearly foreseen. In order to make an accurate prediction, appropriate algorithms with desired input and output data must be trained. But if the output data is not available, a switch must be made to other methods. In such cases, the clustering methods of unsupervised learning can be used. Clustering is used to find patterns and groupings in data. No predefined output data is needed to find a subdivision.
2- Preventative maintenance
One of the main reasons why preventative maintenance could be useful is the capability for predicting when a
mechanical part may require replacing. Combined with historical evidence, machine learning produces an algorithm that detects possible problems when they emerge, helping organizational specialists to take the steps required to eliminate problems that can delay or even interrupt development.
In preventive and predictive maintenance, statistical methods have been used for some time in making decisions. In areas with many variables, machine learning methods such as neural networks can be used for classification. Angius et al. (2016) have shown, however, that the policies in such systems can only be poorly implemented and may sometimes affect the completion and the delivery dates of customer orders. Therefore, not only the condition of the machines needs to be taken into consideration, but also the impact of these system policies on the service level of the system, before choosing them.
3- Predictive forecasting
To stay competitive in today’s ever-changing economic climate, businesses must stay vigilant even to slight
changes in market patterns, which can suggest substantial variations in demand in the future, since this can cause severe upstream production problems. All elements of the purchasing and supply management systems may be handled by AI algorithms to brace organizations for demand changes that could otherwise affect production and delivery. As Mahya et al. (2020) show, in typical supply chain management problems, it is assumed that demand, cost and capacity are known factors. This, however, very often appears to be wrong, as there are various risks arising from fluctuations in demand from consumers, distribution of products and operational risks. Anticipating the changing market conditions allows businesses to become highly resilient and focus time and resources on the most critical points, adopting a proactive versus reactive strategy. As seen in the work of Tarallo et al. (2019), a strategic consideration for suppliers and retailers is a more detailed predictor of the market for fast-moving consumer products. In sales forecasting for short shelf-life and highly perishable goods, the advantages of applied AI methods exceed the precision level of conventional statistical techniques and as a consequence, boost inventory balance across the chain, minimize stock-out rates, enhance supply and increase profitability. Reinforcement learning algorithms are best suited for such market predictions. In areas such as the stock market or crypto exchanges, trading bots that use deep reinforcement learning have been successful. Reinforcement learning algorithms, in contrast to the other machine learning methods mentioned earlier, do not need as much data to achieve promising results. This approach is much more oriented towards human learning behavior of learning optimal ways through trial and error. The agent, i.e. the learner, needs an environment that the learner can influence with own actions. The environment sends him rewards for his actions. The goal of the agent is to maximize these rewards to find the best way for the environment. Adapted to the market, the required parts for a reinforcement learning approach would be clear. The agent would be the bot, the environment would be the market, and the reward would be the profit
or loss generated.
4- Real-time monitoring
One of the most valuable advantages of AI in manufacturing is real-time monitoring, as it gives a more accurate description of where and if any inefficiencies exist in the production chain and what causes the bottleneck. The potential to identify the exact process that needs adjusting, helps organizations to solve the problem rapidly, resulting in time and cost savings. The benefits illustrated in the work of Kumar et al. (2018), show that cloud manufacturing, a real-time monitoring method, may lead to increased efficiency of resources by recognizing the current machine state, minimizing system downtime with the help of condition-based real-time tracking through analysis of the obtained sensor data. This information can then be reused by machine-to-machine communication protocols and cloud service data retrieval methods. Moreover, this concept can help small and medium-sized companies (SMEs) registered in the network, which can benefit from this cooperation and provide cost-effective production services with short lead times.
5- Supply chain management
Machine learning (ML) systems and neural networks can also be extremely useful in supply chain management. In this context, algorithms such as linear regression can be used to predict the impact of the bull-whip effect. Decision trees and random forest can be utilized to perform lead scoring for supply chain managers to allocate resources. Neural networks may be put to use in supply chain management to analyze customer-seller audio and video communications and to plan and adjust lead time. Machine learning is in general important to optimize the decisionmaking process in the flow of goods and services alongside supply chain management. Properly applied, these methods can lead to time and resource savings. In particular, the planning process can benefit from well-known statistical methods that have long been used and extended by ML. Especially for non-linear problems, ML has a fundamental advantage over more traditional methods. Despite the clear benefits, a study shows that ML with one or more supply chain functions, was applied in only 15% of companies. The lack of data or ignorance about the subject could be reasons why such methods are not yet much more widespread. These technological developments influence the purchasing and supply management function and the personnel deployment and will improve those systems in the future.
The purchasing departments of companies generate huge quantities of data, but unfortunately the great potential behind data volumes on this scale is often not fully exploited. Whether due to lack of resources or understanding, this data is often not fully analyzed and processed. Furthermore, purchasing areas are continuing to develop. Markets are becoming larger, more complex, and more competitive. This is where AI and ML come into play. In decision-making in general, methods that have been used for a long time can be improved through neural networks. A key tool of supervised learning is the so-called decision tree. Decision trees look at as many outcomes as possible and try to find the best solution for each decision. One of the most widely used methods for incorporating AI in purchasing departments, is to automate and optimize processes. These improvements can be achieved with similar techniques, which are used in supply chain management. Globalization has also increased the number of markets. the large range of products makes it difficult for people to keep an overview. This is where neural networks, which classify the offers according to certain features, can be used again to make the purchase decision autonomously or to make it immensely easier. Through this application, companies could develop a non-negligible advantage over competitors who do not use such methods.
The manufacturing sector and purchasing and supply management have the perfect fit for artificial intelligence implementation. While the revolution of Industry 4.0 is still in its early stages, we are already seeing major benefits from AI. This technology is intended to transform forever the way in which we produce goods and manage materials, from the design process and manufacturing shop floor, through to the supply chain and administration. The broad topic of AI in manufacturing and purchasing and supply management must be a core element in the curriculum of higher education institutions for all technical fields of study to be future-oriented and to become even more so.