“Robotics process automation” makes a reference to automation. On the factory floor, automation is a well established essential function. We consider the vast experience in industrial automation helpful for a more in-depth view on automation in general, beyond robots, and will look into possible further developments in hyperautomation based on industrial automation established practices.
The new terms “Robotic Process Automation (RPA)” and “Hyperautomation (HyA)” both tackle the problem of how to use Digitalization to improve efficiency of human-executed business processes.
Business processes as well as manufacturing processes are represented as connected work steps. When more formally modeling the process, the connection between steps represents an information flow, or, in the case of factories, the flow of material. Business Processes collect, combine and transform information until a desired result is achieved. Factory Processes collect, transform and combine material and software until a desired product is created. Factory Process Automation is a rich discipline which has contributed significantly to decreasing costs, increasing quality and transforming economies by making mass production and now mass customization possible. Business Process Automation is relatively new, and approaches such as RPA and HyA do not even address the Automation & optimization of the entire processes, but rather just replicate human-executed workflows digitally. In this paper we explore how optimization and innovation techniques used in factory automation may be applied to BPA to go beyond the promises of RPA and HyA.
Looking at the topic from industrial automation, a field with more of 50 years of history, led us to make the link between the two, RPA on one side, and industrial automation on the other side.
Automation in factories – An Overview
Understanding the Factory Structure
Industrial production sites can largely be divided into two types of production schemes: process plants, and discrete factories. Process plants continuously produce something, mostly 24/7. Process plants may be producing oil, gas, chemicals, power, cement, etc. Process plant output is typically measured in volume or weight. Discrete factories (we will use “factory” for short in this text) produce pieces of product, for example cars, consumer electronics, industrial machinery etc. The output of a discrete factory is measured in pieces produced. Due to the closer similarity to business processes, we will start our analysis in the context of a discrete factory and draw analogies from there.
Seen from a high level of abstraction, a factory is taking material (raw material or pre-fabricated components) as an input, and transforms it into the finished good, which in itself can be a pre-fabricated component for a more complex product or system (e.g. motors, computer chips).
A factory comprises the following elements:
- The objects to be produced (product)
- Material and components that the product is made of
- Machinery to transform material and components into product, also robots
- Input-, output-, and intermediate storage (warehouse, shelf, etc.)
- Transport elements (conveyor belts, vehicles, cranes, pipes)
- Sensing elements from simple sensors that indicate whether a part is in position, to more complex quality control
- Human workers and operators
Some production steps, often material transformation (drilling, bending, etc.) are performed in machines. These machines perform a function mostly automated (machine automation). Between machines, the components are transported (automatically or manually) to other machines, or to intermediate storage. A series of machines and production steps form a production line that very often ends in assembly, where robots or humans are putting the components together.
The automation level within a machine is typically high. Along the production line, automation level may vary: fully automated robot cells and manual interventions may be chosen depending on the complexity of the task. The sensory equipment along the line provides the information necessary to confirm the correctness of the execution.
A product may need to go through a number of production lines in sequence, mostly with storage elements in between. There may be parallel lines that produce the same, or similar products within the same factory.
Digital factory automation
As already mentioned, the individual process steps are digitally controlled, using automation elements such as programmable logic controllers (PLC). These components are connected to sensors and actuators and control the proper execution of the steps.
The function of the factory is defined through the software that is programmed into the PLC. When products are changed, PLCs may have to be re-programmed, or reconfigured. For this purpose, they have to be connected to the engineering environment where the products are defined.
In many factories, the data is collected centrally and the performance of the factory is monitored through a manufacturing execution system (MES). This system controls produced batches of products, and supervises the production process. It is the digital window into the plant.
Business process integration
Within the enterprise, the factory is part of the overall business processes. As mentioned before, the factory must receive the information on what products to produce. If produced to order, this is linked to a customer specification that was received as an order, with corresponding sales and order handling processes before.
The production process in the factory can therefore be seen as a part of the business process that involves physical production steps, but it is embedded in a business pre-process, and also in a post-process (distribution and delivery). The business processes as well as the production processes are interacting with the ecosystem around the enterprise: orders are received from customers, orders are sent in turn to suppliers, who then ship material that is transformed within the factory, and the final products are shipped to the customer, who then pays for them.
The whole ecosystem works by delivering physical components / products (manufacturing process) and receiving money in turn (business process). Hence, the business and manufacturing process are not only linked within the factory, but also along the value chain.
From Factory automation to Business Process Automation
As we have seen in the previous section, business process and manufacturing processes are two sides of the same coin: producing and delivering products that are paid for (at this point we exclude service related business models).
This leads to the hypothesis, that similar structures can be used to automate both the factory, and the business process. However, we need to do a mapping of the entities on an appropriate level of abstraction.
Within the factory, the object that is transformed from material and pre-manufactured is the tangible product to be manufactured. Material and parts can be tracked from inbound logistics, through warehouses, machines, to the final product. Each product is an individual entity that can be uniquely identified at all times, as it is a tangible object that has a unique location within the factory, and is in a defined state at any stage of the manufacturing process.
As a first layer of processing we have the machines that do the material transformation and assembly. These are the devices that perform the transformation.
These devices and machines contain their own automation algorithms that instruct them how to perform a particular step in the manufacturing line, e.g. welding two parts, or assembling components with robots. The complete factory is coordinated by factory automation, the algorithms that coordinate the behavior of the automated devices, robots and machine, and that observe the proper handling of the parts between machines along the production line.
If we try to map these components: manufactured object, machines, and automation, we can draw the following analogies:
The factory automation level that coordinates the proper execution of the production line is a description of the production process. Similarly, in business processes, this is the process description. We will later investigate how the automation principles can be transferred from the factory floor to the business layer. If we conclude from the similarity of the process description, the manufactured object and the machinery can be identified as follows:
The manufactured object is a set of information that is handled through a business process. This can be a customer order, an employee onboarding, or any other information subject to the processes of an enterprise.
The algorithms that transform the information can be related to the machines that transform the material. A customer order is matched with the customer’s relationship information (address, credit rating, discount level, etc.), i.e. the algorithm has to add further information to properly execute the customer order. Similar to a machine that is adding screws and bolts that are consumed from a warehouse according to a bill of material (BOM), the algorithm can pull further information from repositories according to the underlying business processes, and can transform the information in the customer order to generate the information that is required to execute the order.
In the case of RPA, software is used to copy a process executed by a human. The value achieved by implementing RPA is simply the saving of effort (reducing personnel) and increasing the speed of execution. In the factory this would be equivalent to just copying the actions of a laborer. This implies that it is only applicable to simple, repetitive tasks where there are few exceptions. Any exception handling or complex tasks will need to be handled by humans, eliminating the benefits.
In contrast, factory automation and optimization re-engineers processes – both the steps being automated and the information flow are rigorously analysed and re-engineered to ensure that they can cover any exceptions and require the absolute minimum of interaction. The KPIs are not only saving effort and increasing speed, but also improving the quality, which is left constant by RPA. In the next section we explore how such optimizations are implemented in factory automation, and how it may be transferred to business processes to reach the same benefits.
What have we learned?
What have we learned from automation, optimization and control of factories over the years? (from application to RPA)
In the following sections we explore two levels of application of the ideas from factory automation to the problem of business process improvement. The first level deals with optimization techniques, which already go beyond the current practice of RPA as they involve changing the underlying process. The second level applies advanced techniques based on advanced mathematical techniques to radically change the approach to understanding and optimizing processes. This is the level of vision, and fully exploits the potential suggested by drawing equivalences between factory automation and business processes.
The first level of optimization may be applied by applying heuristic techniques – also known as common sense. We do not claim that these are fundamentally new, indeed they are certain to be applied elsewhere, but we include them as a starting point for the following development.
Heuristic techniques involve analyzing individual process steps and information transfer, and then re-organizing them to remove duplication and unnecessary re-work. Examples would be performing the same text processing in different tools, or noticing that several steps all require the same input information, but that they are coming from different sources. Another example to detect duplication is to follow the business object being transformed, and to see if the same information is entered more than once. Here centralization of the data or a re-organization of the information flow may be applied. By taking a birds-eye view of the process, loops and repetition may be identified and removed.
The second step is to apply ideas which have arisen from improving the quality control of the factory automation process. In a factory it is clear that errors introduced early in the process can have a significant impact on the final process quality. Therefore quality checks are built in early and often in the process. Such processes are even easier to introduce for business processes which are based on information flows. Starting at the data input, plausibility and data type checks can be introduced. At each step in the further processing of the information / business object, quality and plausibility checks can be built in. This ensures a better overall result due to the lack of rework and there now being no need to correct earlier mistakes.
In factories, physical objects are processed, and these processes need time. Since not all machines operate at the same speed, this leads to the need to introduce storage between process steps, which is wasteful. In order to address this, network analysis has been developed in order to identify the process steps which require the most time, or identify bottlenecks. By re-engineering these points of the process, the overall process may be sped up and optimized. Through digitalization, automated business processes should not have this problem. However, the execution still requires resources, or may produce waste. A network analysis of the overall process will allow identification of the steps which are holding up the overall process, and then re-working these steps will improve the overall process.
Finally, a thorough analysis based on Systems Engineering may be applied to re-architect and optimize the overall system of processes. In Factory Automation, sub-processes or entire processes may be re-designed by considering them as a single system, with a given set of inputs (Material, Information), resource requirements, and then a final result (product, quality, total time taken). By reconsidering what enters and leaves the system, new ideas for the processing within the system may be developed – optimizing the resource requirements and the final result (quantity, quality, time required). This approach may also be applied to business processes. The most simple example of this sort of re-engineering consists of consolidating a series of process steps and then re-engineering them to be a single process step. For more complex systems, the following questions are of particular interest: What are the interfaces to the external world (customers, suppliers, partners)? What are the measures of quality in the result? Can the outcomes and the resources necessary to produce them be quantified? In particular the latter steps enables the use of optimization techniques to derive optimal processes.
The Vision of Factory Automation
If we want to look beyond RPA, we need to have a better understanding of the overall process. The basis for this is a model of the overall information flow. Once the process has been automated, the flow of information can be analysed. Bundling common information flow may help to not only identify redundancies, but also to re-group processes according to their information needs. This approach to re-engineering the system is based on the idea of minimizing the information flow & thus the interdependencies between different process sub-groups. This can ultimately lead to optimization when implementing the processes in a distributed system, or when designing an information security architecture for the overall system. Fewer interdependencies can also contribute to the resilience of the overall system.
A more radical approach to a full (systems based) process re-engineering involves moving in the opposite direction. Instead of aggregating processes as much as possible, individual process steps are dis-aggregated into small steps, each with few inputs and outputs. An overall supervisory system can then analyse the available information and choose the possible steps in order to move the process forwards. This enables more robust processes – which are potentially able to cope with missing information, or more interactive approaches, as the supervisory system identifies key information deficits and alerts the human operators so that they can provide the necessary information. In this vision, there are no longer fixed processes, but rather many parallel processing threads, each of which is moved forward as fast as is possible with the available information. The theory for such an approach lie in the ideas of Moving Horizon Optimization and Game Theory. The system output is formulated as an optimization problem for the next period, and is iteratively solved based on the available information.
Interestingly, this approach is the mathematical equivalent of what a team manager does. Continually analyzing the work and output of his workers, a good manager will dynamically assign work to optimally utilize the skills and times of his employees to maximize the company KPIs.
Piqued by the use of the term “Automation” in RPA and HyA, we have explored the analogies between business and factory processes, and developed new approaches rooted in Industrial Control & Automation which may be applied in the area of business process optimization. As we have demonstrated, there is considerable potential to go further than the current scope of RPA and HyA. A more detailed application of these ideas in a real business context would quickly prove the feasibility or need for further development.
Authors: Christopher Ganz and Andrew Paice