Blog: Manufacturing Resilience
Towards Resilience Management in Production Engineering
The term Resilience became popular in the seventies and generally describes the ability to respond to disturbances. In different disciplines, resilience is used more or less uniformly, although the term disturbance can have far-reaching meanings.
- In Psychology, resilience is the ability to cope with critical situations or to return to a status prior to this situation quickly. Resilience exists when the person uses (mental) processes, methods and behaviors in protecting self from the potential negative effects of Stressors. In simpler terms, psychological resilience exists in people who develop capabilities that allow them to remain calm during crises/chaos and to move on from the incident without long-term negative consequences.
- In Ecology, resilience is the capacity of an ecosystem to respond to a disturbance by resisting damage and recovering quickly. Such disturbances can include stochastic events (such as fires, storms, insect population explosions,…) and human activities (such as clearing of woods, emissions, fracking of soil to extract oil,…).
- In Engineering Sciences resilience refers to the ability of technical systems not to fail completely in the event of faults or partial failures, but to maintain essential system services. Technical systems are called resilient, if they maintain the requested system achievements also with the occurrence of internal and external failures and disturbances.
In this context Manufacturing Resilience describes the ability to manufacture a constant quality at reasonable costs and production time despite disturbances and uncertainties.
2. Quality Issues in Production
The goal of every production is a benefit-giving and customer-oriented manufacturing. This means that quality must meet customer needs and costs as well as production time must be kept to a minimum. If a production is not resilient, disturbances can have serious consequences for the quality, the costs and the production time.
There are 8 types of waste according to MUDA (a Japanese word meaning “futility; uselessness; wastefulness”), which from the customer’s point of view do not add any value to the product and are favoured by poor resilience management.
- Transportation: Every time a product is touched or moved unnecessarily there is a risk that it could be damaged, lost, delayed, etc. as well as being a cost for no added value. Transportation does not add value to the product, i.e. is not a transformation for which the consumer is willing to pay.
- Inventory: Whether in the form of raw materials, work-in-progress (WIP), or finished goods, inventory represents a capital outlay that cannot yet produce an income. The longer a product sits in one of these states, the more it contributes to waste. The smooth, continuous flow of work through each process ensures excess amounts of inventory are minimized.
- Motion: In contrast to transportation, which refers to damage and transaction costs associated with moving the product, motion refers to the damage and costs inflicted on what creates the product. This can include wear and tear for equipment, repetitive strain injuries for workers, or unnecessary downtime.
- Waiting: Whenever the product is not in transportation or being processed, it is waiting (typically in a queue). In traditional processes, a large part of an individual product’s life is spent waiting to be worked on.
- Over-production: Making more of a product than is required results in several forms of waste, typically caused by production in large batches. The customer’s needs often change over the time it takes to produce a larger batch. Over-production has been described as the worst kind of waste.
- Over-processing: Doing more to a product than is required by the end-customer results in it taking longer and costing more to produce. This also includes using components that are more precise, complex, expensive or higher quality than absolutely required.
- Defects: Having to discard or rework a product due to earlier defective work or components results in additional cost and delays.
So the task of a Resilience Management in Manufacturing is to flexibly manage resources that allow us to manage, power and move the production system efficiently as a reaction to occuring disturbances, while keeping the losses or waste as low as possible.
3. Resilience in Manufacturing
A value chain is a highly complex system. Raw materials are extracted from soil or above ground, enriched into useful materials, which in turn can be used to manufacture semi-finished products. Several semi-finished products become components, components become assemblies and assemblies become products. Each of these steps is highly complex and has different far-reaching consequences or requirements for production. Nevertheless, we want to define a value chain by 3 principal levels: micro level, meso level, and macro level.
Micro — From Raw Material to Components
Real world example: Fine blanking is a high sophisticated and very complex process. Up to 250 fine blanked parts are used in automobiles of every size. Researchers of WZL in combination with fine blanking specialists have calculated a loss of over EUR 1.3 million per year at a waste level of only 2.5 % for a typical stainless steel car component [BECK19].
The production of semi-finished products and components is usually carried out with machines that cover all or most of the production steps. Although these machines are highly automated, statistically no identical semi-finished products or components are manufactured. On the one hand, the reasons lie in nature: materials are natural products which are obtained from raw materials. Naturally, they are not perfect and not idealistically uniform. Logically, this results in deviations in production if the production parameters are not adjusted. Due to a lack of Resilience Management, this ability is currently only possible with experienced personnel.
Nevertheless, there are many other disturbance variables. In addition to the material, the attention of the machine operator is subject to fluctuations, the power grid operates the machines with different intensities, the outside temperature changes over the day and thus also the temperature in the shop floor. Even if all technical properties run perfectly, errors can still occur in the analysis, evaluation, and assembly of the components.
For this reason, tolerances have been introduced for technical products that make it possible to produce economically within a valid spectrum. The task of the Resilience Management is therefore to enable economic production without loss of time within the required tolerances in case of major disturbances by reorganizing manufacturing.
Meso — From Component to Products
Real world example: According to Statista, the economic loss due to health-related, like flu epedemics and more, production losses in the manufacturing industry in Germany amounted to more than EUR 18 billion in 2013 [STAT19].
The production of a product usually involves more than just one work step. An average passenger car today consists of up to 10,000 individual parts. Depending on the size and equipment of the vehicle, it can also be more. This means that the susceptibility of the system to faults is much higher than on a micro level alone.
Errors in production lines can be inherited, but can also be fixed. However, delays on one line can also lead to delays on another line. The task of Resilience Management is therefore to manage the intralogistics of a company. But the organisation of personnel is particularly demanding. The manufacture and assembly of products with up to 10,000 individual parts requires a large number of highly trained employees whose interaction is highly complex. If one or more of them is unable to work due to illness, his knowledge and skills will also be lost. Therefore, the complexity of these systems needs to be broken down so that worker with less experience can handle the process as well, in case the specialized personnel is unavailable. A tool to deal optimally with this issue are explainable AI appraoches, which will also be part of reseach in context of Resilience Management.
Macro — Outside the Firm
Real world example: The past has shown that supply bottlenecks at one supplier can lead to costs of at least EUR 100 million per week [WELT16]; bottlenecks at two suppliers to costs of at least EUR 410 million per week [ZEIT16]. The damage to the economy is even greater, as employees are sent on forced leave or short-time work is enforced. In the above example, up to 60,000 employees were affected. [STAT19].
However, the purchase of raw materials or the sale of production is determined by more than just efficient production. Politics and economics determine most of the rules of production at the macro scale. Trade prices rise, goods become short, political debates lead to customs duties and uncertainties (see Brexit debate 2019).
The task of Resilience Management is to find an optimal solution despite far-reaching uncertainties, so that companies ideally emerge unweakened from any event, e.g. by anticipation of supply bottlenecks, by moving river routes to air or road routes in hot summers, etc. This requires an AI system that can grasp and process the most diverse levels. In this context, the project relies on the principle of the digital shadow from the Internet of Production and combines it with the Edge AI principle.
4. How Edge AI might help!
AI requires a wide variety of data sets in order to be able to recognize essential relationships. The cloud system makes such a variety possible. Data can be aggregated worldwide. However, the disadvantage of the cloud is that only strongly reduced structured data records with high latency can be uploaded. This may be sufficient for many business processes, but especially at the manufacturing level, where machines — weighing several tons and generating data streams of several Gbit/s— are controlled within milliseconds. Thus AI algorithms must work in approximate real time. The layer, where AI is applied locally in devices near the manufacturing machines, is called Edge. Edge AI reliably distributes and relates low-latency AI computations on devices while supporting data protection and data control at low resource consumption.
Edge AI as an addition to cloud systems needs to fulfill the following requirements: Local perception of sensory data by embedded sensors, Local persistent primary data storage, Complex data analysis on local devices, Backup with trusted third party data storage provider, Role-based access-controlled exchange of data between local systems, Role-based access-controlled peer-to-peer exchange of data, models, and software components between local systems on different machines, and resistance against intrusion attacks and data fishing. However, since operational technology has grown historically, at a time when networking was not possible as it is today, each manufacturer has more or less developed its own quasi standard for machines. These standards are not necessarily the best, but they are widely used in the field and cannot be replaced immediately. Therefore, the most critical aspect for Edge AI is the creation of a unified standard or translator protocol to connect the different machines, e.g. with OPC UA.
With Edge AI in mind, three use cases in manufacturing are defined, which should show the potentials and challenges of Resilience Management. Edge AI will be applied in the micro level, a hybrid Cloud & Edge AI at meso level, and Cloud AI at Macro level.
5. Use Cases
This section is based on the research project #SPAICER, which considers the fine blanking press known at the WZL as a use case. Fine blanking is a metal cutting process designed for mass production, e.g. for the manufacturing of brake calliper carriers or belt straps. Fine blanked components thus often perform safety-relevant tasks, e.g. in automobiles. The following video shows the fine blanking machine, consisting of a decoiler, a leveler, a lubrication unit, and the press, as well as the press’ process kinematics during blanking operation.
Based on the distinction between micro, meso and macro, three use cases were defined in #SPAICER.
Use Case 1: Micro level — Material variations
At micro level, material variations and uncertainties in the processing of the material are considered. The responsible decision circle is focused on the machine operators. Their primary task is to keep things running, to get the best out of the machine and stick to the schedule.
Use Case 2: Meso level — Unforeseen losses and absence of workers
At the meso level, fluctuations and absence of well trained workers in human resources are considered. The affected decision circle is designed for middle management. Their primary task is to keep quality as high as possible and to find ways to compensate the loss of perfectly trained workers.
Use Case 3: Macro level — Supply bottlenecks
At the macro level, supply bottlenecks are simulated which can influence not only plants but entire groups. This is where top management is addressed. Supply bottlenecks will have significant impact on the entire firm. Top management needs to rethink processes, reallocate resources and keep the financial losses as low as possible.
The use cases and real life examples show that a resilience management in manufacturing in combination with Edge AI approaches has great potential to make an economic contribution that is attractive not only to the entrepreneur, but also to people and the society.
[BECK19] Beckers, A.: Zwischentreffen-Report Arbeitskreis Feinschneiden 2019.