Automation architecture for harnessing the demand response potential of aqueous parts cleaning machines

We define our requirements for analysing industrial DR potential in order to evaluate how existing methods can be applied to aqueous parts cleaning machines:We use these requirements to analyse existing methods for potential analysis.

In the following, we present a brief literature review of existing methods to determine the DR potential in the industry sector. We sort the methods into two main categories: Determination of the potential of the factory as a whole and of individual production machines. As far as we know, there are no methods for the analysis of cleaning machines’ DR potential in particular.

Most research focuses on evaluating the entire factory: One method is to map the factory as a simulation model and estimate the DR potential using simulation [15‐17]. Other research attempts to assesses the DR potential based on DR measures [18, 19]. A different approach to examine factories is the evaluation via key figures and estimation of the potential based on axioms and the energy flows of the production site [8, 9]. Furthermore, a comprehensive method of analysing the whole factory and its processes is presented in [10]. The procedure includes an analysis at module level as well as the systematic development and characterisation of suitable DR measures. Another easier-to-use method is described in an industrial guideline by the Association of German Engineers e.V. [20]. Here, the suitability of entire machines is estimated and compared with that of individual machine modules. After an initial evaluation, more detailed energy measurements have to be carried out subsequently.

In contrast to the previously mentioned studies, the following research focuses exclusively on the machine level. [11] shows a method for the preselection of relevant modules regarding their suitability for defined measures which results in EF characteristic diagrams. In [12], the EF of machine tools in the production state is investigated. The approach is based on individual estimations for each module. [13] extends the approach of [12] with the methods from [8, 11] and also applies it to machine tools. This method does not require the prior definition of specific measures and results in a quantified value for the existing technical energy potential.

In addition to the above-mentioned methods for analysing the industrial DR potential, the VDI guideline 5207 Energy-flexible factory - Fundamentals should be mentioned. Here the authors define common DR measures for the whole production site including the machine level in detail [6].

To estimate the DR potential of aqueous parts cleaning machines we adapt existing findings to aqueous parts cleaning machines and extend them where necessary. We use the VDI list of common DR measures as a basis for the selection of suitable DR measures for aqueous parts cleaning machines. Then, we select suitable methods for calculating the DR potential of these measures.

The VDI 5207 guideline presents an overview of DR measures in production sites. The measures are divided according to the automation hierarchy into manufacturing level, manufacturing control level and enterprise control level. Due to the aforementioned requirements, only the following measures on the manufacturing level are relevant [6]:As the first process step could be interrupted before its start, we also consider the measure shift start of job, even though it is located on manufacturing control level. In the following, we analyse it as part of interrupt process.

Not all DR measures are applicable to cleaning machines. The energetically relevant cleaning processes, which usually have the highest energy consumption, are cleaning and drying. Logically, the order of these cannot be changed, excluding the measure change process sequence from further consideration. In principle, it would be possible to adjust the process parameters of a cleaning machine. This could, however, affect the cleanliness and/or dryness of the part as a result, which is why we don’t consider the measure adjust process parameters. We also omit the measure operate with bivalent energy, since we restrict ourselves to an electrically supplied cleaning machine. Therefore, we only consider interrupt process, including shift start of job if the interruption is before the first process step, and store energy (inherently) as DR measures in the case of an electrically supplied cleaning machine. The selected measures require potential analyses at machine component and at process level. For this purpose, we examine the existing methods for potential analysis.

Since we want to estimate the DR potential in order to have a preselection of the modules to be used for DR measures, in our case simulations are unsuitable due to the high effort needed for modelling. Furthermore, most of the approaches presented in Sect. 2.2 that take into account the factory as a whole [8, 9] are unsuitable, as we focus on the machine level. In other approaches [10, 18], specific DR measures have to be defined in detail prior to the potential estimation, including exact specification of lead time and call-off duration. This requires detailed knowledge about the use case and is too extensive for a preselection. Therefore, we choose the method for machine tools presented in [13] as a basic approach for potential analysis of the modules and the process. This approach is located at machine level and includes aspects for analysing machines from [12] and [19]. It is also partly similar to the VDI guideline [20]. However, only machine modules are considered instead of the whole factory. Thus, the evaluation criteria are more suitable for the analysis of machine modules.

The potential analysis is divided into two parts: First, the machine is analysed with a focus on its electrical consumers in Sect. 2.3.1. In this step, we identify the inherent technical DR potential. Second, we examine the cleaning process as a whole as well as its individual process steps with regard to their suitability for an interruption of a running process step in Sect. 2.3.2. Both steps combined form our method for the DR potential analysis of aqueous parts cleaning machines. An overview of the approach is depicted in Fig. 1.

Up to now the, industrial automation architecture has been structured according to the automation pyramid defined in IEC 62264 [21], see Fig. 4. Production machines and their modules are part of the field and the control levels. The machines’ sensors and actuators are located at the field level. They are connected to and controlled by programmable logic controllers (PLCs) at the control level. The control level is connected to high level systems such as supervisory control and data acquisition (SCADA) systems, manufacturing execution systems (MESs) and enterprise resource planning (ERP) systems which are located on higher levels. The automation pyramid ends at the company boundary and is not designed to communicate with entities outside the company such as a DR market [22]. In [22] the automation pyramid is extended by three additional levels to include the energy market. However, for the energy-flexible usage of production machines it is inconvenient to use the proposed automation architecture that consists of multiple levels: As requests from outside the business unit are brought in via the ERP system, all other levels of the pyramid must then be traversed. This can lead to a very slow and complicated processing of DR measures. Therefore, we present an alternative concept in this paper.

To calculate DR measures, often complex optimisation procedures are used [23, 24]. These optimisation algorithms are usually provided in high-level languages such as C++, Java or Python and are executed on personal computers (PCs) on the information technology (IT) level [25]. PLCs which execute the DR measures are located on the operating technology (OT) level and their automation program is usually written in IEC 61131-3 languages [26]. The PLC automation programs on OT level are designed for fast and simple automation tasks that are executed in real-time but have limited capabilities for the implementation of complex algorithms. Optimisation algorithms often need a few seconds for their calculations whereas typical PLC tasks are executed in a guaranteed fixed cycle time of a few milliseconds. Therefore, the algorithms have to be deployed on the IT level separate from the automation program on the OT level.

We identified different approaches for the coordination of DR measures: First, the single machine can be part of an external electricity market or is controlled directly by the grid operator. Here, systems outside the company compute DR measures and the machine is connected directly to these external systems [22]. Second, the DR measures are calculated and executed for single machines or the whole factory by a central system that is located inside the company [27, 28]. Third, the cleaning machine can be part of a decentralised multi agent system where machines interact with each other to find a global solution near the global optimum [29, 30]. The traditional automation architecture is not designed for any of these machine interactions.

Due to the aforementioned factors, we believe that a new automation architecture design is needed for applications such as the energy-flexible operation of machines. One possible solution is to model the automation as a cyberphysical production system (CPPS). Cyber-physical systems are defined as computer systems which interact with other computational entities and are connected to the physical, mechanical world [31]. When applied to production and automation systems they are called CPPS [32]. This concept breaks the rigid automation hierarchy as CPPSs mainly consist of two main functional components: one high level area where more complex functions are implemented and a low level area with functions responsible for real-time data processing and control of the physical machine [32]. Both levels are connected by a third communication component that enables the link between the digital (cyber) elements and the physical elements of the CPPS.

In [33], a CPPS design called automation diabolo is proposed. It consists of three parts:as shown in Fig. 5. In this design, the classical management systems classical management systems SCADA, MES and ERP are located at the top level and connected directly to the field and control level via a new communication layer at the centre. This has the advantage that management functions can address control functions simultaneously and without intermediate steps, thus enabling faster data flow and process execution.

Also, new services, including a DR service, can easily be added to the management functions. The services at the management level are usually executed in classic IT systems. These can be installed externally in the network or they can be part of the cleaning machine as an embedded or industrial PC. The services at the field and control level are usually executed in OT systems such as PLCs.

For the implementation of the communication layer, communication standards that can be understood by IT and OT systems must be used. In OT systems, simple electric binary or analog signals and fieldbus communication protocols such as Modbus or Profibus are used for the interaction between sensors, actors and the PLC. Communication between different IT systems is widely based on the Transmission Control Protocol/Internet Protocol (TCP/IP).

The usage of a service-oriented automation architecture for DR measures is already discussed in literature. In one concept, DR services are included in two interacting IT platforms [27, 28]:Most of the interaction and calculation is carried out at the management level. This is only possible by using a service-oriented automation architecture.

The usage of a service-oriented automation architecture for DR measures in building automation is shown in [35]: Here, the authors focus on the modular object-based structure of the OT automation program that creates a hierarchical automation data model in the communication layer using Open Platform Communications Unified Architecture (OPC UA). The building automation program can communicate with a DR service located at the IT management level via the communication layer. The DR service optimises the operation of the industrial supply system controlled by the building automation based on a changing energy price. The authors emphasise the importance of guaranteeing safe operation of the supply system when it is controlled by DR services. Therefore, they include a safety function that interrupts control by DR services if safety limits are exceeded and activates the standard process control at the field and control level instead. The standard control controls the supply system until a safe state is reached, then it returns the control to the DR services again. An example is to keep the temperature in a thermal hot water storage within the safety limits and to prohibit external access in case the limits are exceeded.

In [2], the authors present a DR service that calculates the IES potential of hysteresis controlled production systems. The service is executed on an IT system at the management level. To calculate the potential, detailed information from the field level, such as the power demand or the current operating point of the machine, is required. Therefore, a connection to the field level is necessary.

In the following, we describe our CPPS approach focusing on the interaction between services at the management and field levels and the communication layer used for this use case.

For the energy-flexible operation of aqueous parts cleaning machines we define the following requirements for our CPPS: We propose a CPPS based on the automation diabolo [33] that takes all described requirements into account, visualised in Fig. 6. For energy-flexible production we add the DR services to the management level. The DR service decides if, when and how a DR measure is to be carried out. In the management level, additional services could be included for example a SCADA service that visualises actual machine states or an energy monitoring service that stores historic energy data in a database.

The communication layer connects the DR service at the management level to the cleaning machine and enables IT-OT-interaction and inter-plant-communication. Here, the information that is needed for energy-flexible operation of the cleaning machine is transmitted. Also, additional data points could be included here depending on the specific application, for example to provide information for additional services such as condition and energy monitoring. For the interaction between the DR service and the machine, we propose using an automation data model that includes information for the energy-flexible control of individual machine modules (store energy inherently) and the interruption of the cleaning process (interrupt process). In the following, we describe the data points that should be included in the automation data model for the interaction with DR services.

Direct control of individual machine modules for storing energy inherently requires information about the current status of these modules. Also, access to the module’s control commands has to be established. Therefore, we include the following information for every module that can be used for storing energy inherently:Some of this information is static and does not change during machine operation and other data points may vary dynamically. Most of the data is accessed in read-only mode except for the DR set point which is written by the DR service. An overview of the data points for storing energy inherently, their static or dynamic type and access mode is shown in Table 1

For interrupt process, information about the process status and the entire machine must be exchanged:All these data points show dynamic machine information. As in the measure store energy inherently, most of the data is accessed in read only mode except for the interruption command. The overview of the data points for interrupt process, their static or dynamic type and access mode is shown in Table 2.

The cleaning machine’s automation program is located at the control level. We suggest to use a modular object-based design when implementing the automation program since it enables easy extensibility and transferability. We include the functionalities process control, energy control and safety control:After presenting our automation architecture for the CPPS cleaning machine we apply it to an exemplary cleaning machine in the following.

The examined machine, MAFAC KEA, is an aqueous parts cleaning machine with a closed cleaning chamber. The central units are the cleaning chamber with its holding basket and spray nozzle system and a 320 ls tank filled with aqueous detergent. Loading and unloading with parts is done manually. The tank temperature control and the air heating are carried out by electric heaters. The cleaning pump supplies the cleaning chamber with detergent from the tank. An oil separation system separates oil residue from the detergent. Air can be supplied or extracted from the chamber via two fans. The holding basket and the spraying frame are driven by electric motors. The total rated power of the machine \(P_\text {c}\) is 20.7 kVA. The installed power analyser, Janitza UMG 96 RM-P, measures the machine’s total power consumption.

The examined process takes 12 min in total. The process steps and the involved modules are presented in Table 3. The first process step spray cleaning is the main cleaning process which lasts for 600 s. Here, the detergent is pressurised by the cleaning pump and sprayed onto the parts. Furthermore, the holding basket and spraying frame counterrotate. The three modules, cleaning pump and the two motors for spraying frame and holding basket, are active during the whole step. In the next step, impulse blowing, the detergent residues which are now contaminated with soil are removed by impulse blowing. For this purpose, compressed air alternately passes through the nozzle system for 10 s. This is followed by a 10 s pause and then another 10 s of compressed air. The basket rotates for the entire 30 s. In addition, the moist air inside the cleaning chamber is extracted by the exhaust fan throughout the whole duration of this step. Finally, components are dried with hot air during convection drying. The air is heated using temperature-controlled electric heaters, which are switched on and off according to a temperature control. On average over multiple process cycles, they are switched on five to six times in the process step, resulting in a total activation time of 57 s. The supply fan is active during the entire process step. In addition, the exhaust fan continues to extract moist air. This last step takes 90 s until the basket finally turns to its starting position and the chamber door opens. During the whole cleaning process the temperature of the aqueous detergent must be kept around 60 \(^\circ \hbox {C}\) and the tank heater is controlled in a hysteresis between 58 \(^\circ \hbox {C}\) and 62 \(^\circ \hbox {C}\). Thus, he tank heater’s activation time is not constant in every process cycle, in some cycles it is not activated at all. We estimated an average activation time of 152 s over multiple cleaning process cycles. The oil separation system is also operating independent of the process steps. It is turned on and off depending on the measured level of oil residue in the detergent. The average activation time is 100 s.

We start with the examination of the modules by looking up the respective nominal power \(P_{\text {n},i}\) of the electrical consumers from the machine’s data sheets and determining the running time of the individual modules, see Table 4. The running time of the modules is defined in the machine control program. Exceptions are tank heater and air heater which are both temperature-controlled. For these modules, we record the activation time over several cleanings and calculate the average activation time. Using the running times, we calculate the energy demand \(W_i\) and share of energy demand \(\phi _s\) for each electrical consumer i for the whole cleaning process using (1) and (2). We only consider controllable consumers that can be used for DR, therefore parts such as the machine’s PLC or the control cabinet cooling are excluded. The results are shown in Table 4.

Based on the nominal load \(P_{\text {n},i}\) and the share of energy demand \(\phi _i\), modules with high energy relevance are selected using two thresholds. As we want to avoid the exclusion of promising modules we choose low thresholds: Only modules with nominal power below 1 kW and share of energy demand below 5% are excluded. The flexibility potential of modules with nominal power and energy demand below these thresholds appears to be too small to be considered for DR measures. Electrical consumers that are at least above one of these thresholds will be considered. The energy demand and share of energy demand of all modules and the thresholds are shown in Fig. 7. The areas below one threshold are coloured in yellow. The area below both thresholds is coloured in red.

The tank heater and the air heater have the highest nominal loads. The cleaning pump accounts for the largest share of energy demand. Exhaust fan, drying fan, oil separation and the motors for basket and nozzle rotation are below both thresholds. Consequently, we do not consider them further, as their impact is too low.

The cleaning pump is only switched on during the step spray cleaning. This corresponds to a process-related control. There is no control of the quantity. Accordingly, the control mode of the cleaning pump is C.3. The cleaning pump supplies the spray nozzles and is directly related to the functionality of the spray cleaning. This is a real-time energy conversion without buffer capacities, which corresponds to mode I.4. The evaluation of energy-process-independence is summarised in Table 4.

Finally, using (4), we estimate the DR potential \(P_{\text {flex}} = 18\,\text {kW}\) including all consumers that have achieved a green or yellow rating and the relative potential \(R_P = 87.0\%\) according to (5). This potential can only be reached if the tank and air heaters are used for DR at the same time. As the air heater is only active for 57 s during the process step convection drying the full potential can only be activated during this short period of time. Since the air heater is averagely switched on six times for 10 s, the full potential is only available for very short-term DR measures such as internal peak load shifting.

After estimating the inherent storage potential, we now consider the potential for interrupting the process. First, for each process step s we examine the relevance for DR by calculating the accumulated nominal power \(P_{\text {n},s}\) of all active consumers using (7) as well as the proportional energy demand \(W_s\) using (8), see Table 5. To cal%culate these values, we use the activation time \(t_{s,i}\) from Table 3 and the respective nominal load \(P_{n,i}\) from Table 4. Equivalently, the limits are set to 1 kW accumulated load and 5% of the total energy demand. The results are shown in Fig. 8. It turns out that spray cleaning accounts for the largest share of the energy demand, while the step convection drying requires the highest nominal power. Impulse blowing is energetically unimportant and therefore not considered further. The total energy consumption of the described cleaning process is \(W_\text {c} = 739.75\,\text {Wh}\). It is important to note that the largest consumer, the tank heater, is independent of the process and therefore not considered.

In the next step, we evaluate whether an interruption of a running process step is technically permissible without negative influence on the process quality. Statements are either based on scientific findings or case-by-case assessment. As already mentioned, a delayed start of the cleaning process has no negative effect on the cleaning result [7]. Accordingly, an interruption before the first step, spray cleaning, is feasible. The second considered step is convection drying. The purpose of the drying process is to remove the remaining detergent from the parts’ surface to prevent corrosion and prepare the parts for sensitive following processes such as hardening or coating. This process is performed after most aqueous cleaning processes and is generally done via evaporation [37]. As far as we know, there is no research on the influence of delayed drying for the cleanliness of the parts. For now, we assume that corrosion is prevented if the interruption is only for a few minutes, making an interruption before the process step convection drying permissible. Regarding the interruption of a running process, we currently cannot evaluate the influence to the cleanliness of the parts. Accordingly, this is avoided until the influence of an interruption on the cleaning quality is known. The evaluation results in a yellow rating for the steps spray cleaning as well as for the convection drying, shown in Table 5.

For the quantification of the DR potential, all process steps with a green or yellow rating are considered. Now we use the energy demand to calculate the absolute and relative DR potential, see (11) and (12). For the KEA this results in \(W_{\text {flex}} = 0.732\,\text {kWh}\) and a relative potential \(R_W = 99\%\).

The potential analysis shows that a direct control of the tank and air heater as well as interruptions before spray cleaning and convection drying have a high DR potential and should be considered for DR measures. We now adapt the control system to enable the machine for executing these DR measures.

In order to implement the CPPS design presented in Sect. 3.2, we have to adapt the existing automation program of the MAFAC KEA located on its PLC. The process control including the field level communication to the machine modules already exists in an object-based structure. We integrate the energy control and the safety control in the existing process control objects. For this, we first partly replace the standard non-energy-flexible controllers that control the machine modules so that the cleaning machine is able to carry out DR measures. Second, we extend the standard process flow control to enable an interruption of the process. Third, we add a separate control function to control the global machine state. In order to prevent negative influences on the process quality and to ensure a safe operation during DR measures, we finally implement safety control functions. In the following, we describe the four steps in detail.

To interrupt the process we add the Boolean variable interrupt and the step interruption to the process flow control function which is part of the process control. If the variable interrupt is true, the program will step into the step interruption until the variable is set back to false. This is only possible at the two decision points marked in Fig. 9.

In addition to including energy control in the module-objects and the process flow control function we add a separate function to the process control that controls the global machine state. Following [36] we define four online states:Depending on the machine state, DR measures can be executed more or less flexibly. In the machine state off, the machine and its PLC are turned off. Therefore, this state is not represented in the process flow control.

We integrate the safety control in the process control module-objects. To ensure safe operation, the direct control of the tank and air heaters is limited to specific temperature limits. The limits for the tank heater depend on the current global machine state. In the states working, operational and powering up, the operating limit is between 55 \(^\circ \hbox {C}\) and 65 \(^\circ \hbox {C}\). The tank heater can be used for DR measures more flexibly during standby. In this case the lower boundary is removed and the higher boundary is set to 85 \(^\circ \hbox {C}\). If the current tank temperature exceeds the acceptable temperature range, the standard process control takes over until a safe temperature is restored.

At the management and organisation level, we implement two exemplary DR algorithms in Python as DR services. The algorithms aim to take advantage of low electricity prices. The first algorithm activates the energy-flexible modules if the electricity price is below a defined threshold. The second algorithm interrupts the cleaning process if the price is over a second defined threshold.

After identifying modules and process steps with a high potential for DR measures for the MAFAC KEA cleaning machine and enabling the machine for energy-flexible control, we now run two field tests to demonstrate the execution of DR measures. As described in Sect. 4.4, the machines automation system is linked to the management level via OPC UA over Ethernet IP. The DR algorithm is executed on a PC that is connected to the same local network as the cleaning machine’s PLC. The PC is a Lenovo Thinkpad T460s with Intel Core i5-6300U 2.40 GHz dual core processor, 16 GB DDR4-RAM and Microsoft Windows 10 Pro 64 bit. First, we execute the DR measure store energy inherently by controlling the machine’s tank heater and second we perform the interrupt process measure by interrupting the cleaning process. For store energy inherently we only consider the tank heater as the energy prices are updated every five minutes and the air heater is only suitable for DR measures with a duration of a few seconds.

The results for store energy inherently are shown in Fig. 10. We run the full cleaning process five times over the duration of two hours. The beginning and end of every cleaning process are marked with dotted lines. The electricity price falls below the defined threshold of 100€/MWh three times, activating the tank heater, as shown in the upper two diagrams of the figure. The three time intervals in which the price is below the threshold are highlighted in gray in all four diagrams. The third interval exhibited a communication problem where the tank heater was switched on with a delay of 70 s.

The third diagram shows the cleaning machine’s power consumption. The total machine power demand rises significantly when the tank heater is activated as it is the machine module with the highest electrical load. When the tank heater is active the first time, the power consumption rises from 2.0 to 10.3 kW, during the second and third time from 2.0 to 12.1 kW. The first activation is during convection drying characterised by its rapidly changing load curve. The cleaning pump is deactivated in this step. The second and third activation occurs in process step spray cleaning where the cleaning pump is active. Therefore, the changes in the power consumption differ between first and second or third activation. The measurement also shows that the actual power consumption of the modules is slightly lower than their nominal load \(P_{\text {n},s}\). The average power consumption of the tank and air heaters are approximately 8 kW each whereas the cleaning pump requires 2 kW. The total power consumption during standby is approximately 0.1 kW.

The diagram at the bottom of Fig. 10 shows the tank temperature during the cleaning processes. The temperature drops by 1–2 \(^\circ \hbox {C}\) each time when the cleaning process is started. The temperature drops become smaller during the experiment. Despite being cooled down in cold water in between process cycles, the cooling seems to be insufficient by the end of the field test. Therefore, future more realistic field tests require more parts to ensure that the machine is loaded with parts at room temperature for each process cycle. During most of the field test, the temperature stays within the safety limits of 55 \(^\circ \hbox {C}\) and 65 \(^\circ \hbox {C}\). Only at the end of the third tank heater activation period, exceeds the upper limit, causing the safety service to intervene.

We execute interrupt process in a second field test, shown in Fig. 11, where we run three cleaning processes. The beginning and end of every cleaning process are marked with dotted lines in the bottom diagram. The electricity price, shown in the top diagram, rises above the threshold of 100 €/MWh twice, triggering a process interruption during these periods, which are highlighted in grey. The process interruption is displayed in the middle diagram.

The bottom diagram shows the total electric power demand of the cleaning machine. The tank heater is active during the first cleaning process and leads to a rise of the power consumption to 12 kW. The process step spray cleaning is running when the process interruption is activated for the first time after 600 s. Since an active process step can not be interrupted, the process is interrupted after 723 s, before the convection drying step. The second interruption is triggered after 2100 s during the third process cycle, postponing its start by 300 s.