1. Introduction
The increasing trend to smart products and sensor-integrating components enables the digitalisation of whole products and digital twins of such (Reference Fett, Turner, Breimann and KirchnerFett et al., 2023). For example, an increasing interest in various microelectronic measurement systems for integrating sensors as additional functions into machine elements could be observed during the last years (Reference Breimann, Fett, Küchenhof, Gomberg, Kirchner, Krause and TrieuBreimann et al., 2023; Kirchner et al., Reference Kirchner, Wallmersperger, Gwosch, Menning, Peters, Breimann, Kraus, Welzbacher, Küchenhof, Krause, Knoll, Otto, Muhammedi, Seltmann, Hasse, Schäfer, Lohrengel, Thielen, Stiemcke, Koch, Ewert, Rosenlöcher, Schlecht, Prokopchuk, Henke, Herbst, Matthiesen, Riehl, Keil, Hofmann, Pape, Konopka, Poll, Steppeler, Ottermann, Dencker, Wurz, Puchtler, Baszenski, Winnertz, Jacobs, Lehmann and Stahl2024; Reference Kruse, Küchenhof, Gomberg, Breimann, Krause, Kirchner and TrieuKruse et al., 2024). When not only small applications like machine components but also large-scale structural components are considered, using a whole sensor network instead of single sensors is required to monitor the large areas. This is, for example, necessary when a component should be monitored regarding damage detection and growth in structural health monitoring applications, for instance, in the field of aerospace (Reference Büchter, Sebastia Saez and SteinwegBüchter et al., 2022; Reference Bergmayr, Höll, Kralovec and SchagerlBergmayr et al., 2023). However, the large geometric scale of components requires a lot of cables for data transmission from sensors far away from the data collection system, which leads to decreased data quality and an increasing weight of the component (Reference Pottie and KaiserPottie and Kaiser, 2000). A possible solution for this problem are wireless sensor networks (WSNs), where sensors are combined to a sensor node at a microcontroller, from which the processed data is transmitted wireless by an integrated transceiver module (Meyer zu Westerhausen et al., Reference Meyer zu Westerhausen, Raveendran, Lauth, Meyer, Rosemann, Wawer, Stauß, Wurst and Lachmayer2024b).
Especially from a product development and design point of view, the data collected by such WSNs is of great interest when it gives insights into the operational loads occurring during a component's use phase. This data enables the optimisation of the components' structure to face the operational loads in the following product generation while, for example, saving weight (Reference Albers, Bursac and RappAlbers et al., 2017). Since the information on operational loads becomes an integral part of a component when the sensors are integrated into the material, this information is “inherited” from one product generation to the next. Therefore, those components are referred to as “gentelligent” in the paradigm of technical inheritance (Reference Lachmayer, Mozgova, Reimche, Colditz, Mroz and GottwaldLachmayer et al., 2014; Reference Lachmayer, Mozgova, in Krause and HeydenLachmayer and Mozgova, 2021). One way to get insights into the load and deformation field of such a gentelligent component is to apply so-called shape sensing techniques from the field of structural health monitoring (SHM). These techniques allow the reconstruction of this field information from discrete measurements, e.g., by using strain gauges, which might require a lot of sensors for good accuracy (Reference Gherlone, Cerracchio and MattoneGherlone et al., 2018; Reference Esposito and GherloneEsposito and Gherlone, 2020). This high sensor quantity leads to the necessity of a WSN. Since in literature the use of a WSN for real-time shape sensing is not emphasised enough, this paper is addressing this gap with a conceptualisation of a WSN for this purpose. Therefore, this paper is structured as follows. Section 2 presents related works on the topic of WSNs and shape sensing techniques. Furthermore, a comparison is conducted to present the research gap for this work. In Section 3, a concept for a WSN and the application of a shape sensing technique on the data collected by it is presented. This proposed concept is then applied to a gentelligent demonstrator in a case study, and the results are presented in Section 4. Finally, Section 5 concludes this paper and gives an outlook on future works.
2. Related works on wireless sensor networks and shape sensing
Different studies on WSNs and their applications, as well as on shape sensing and the application of different techniques, can be found in the literature. To give an overview of each of these research fields and to compare these related works, the following overview is given to derive the research gap for the methodical development of a WSN for shape sensing with gentelligent components.
Shape sensing techniques are based on the use of strain sensors like strain gauges or fibre optical sensors, which measure strains at discrete positions on a component under load and reconstruct the deformations and loads by solving inverse problems from structural mechanics. For this purpose, four groups of shape sensing techniques could be distinguished: (1) methods based on the numerical integration of experimental strains; (2) methods using global or piecewise continuous basis functions to approximate the displacement field; (3) methods employing Artificial Neural Networks (ANN); and (4) methods based on a finite-element discrete variational principle (Reference Gherlone, Cerracchio and MattoneGherlone et al., 2018).
In the first field, Ko's displacement theory is found in literature, which has the advantage of being easy to implement due to simple equations. Besides, it also allows the reconstruction of applied loads due to the relationship between deflections and shear forces and bending moments from solid mechanics. The major disadvantage of this method is the limitation to beam-like structures due to the Euler-Bernoulli beam theory as the basis of it (Reference Ko, Richards and van TranKo et al., 2007). However, the applicability and accuracy of this technique is demonstrated in different studies, e.g. by (Reference Valoriani, Esposito and GherloneValoriani et al., 2022) on an airplane wings. An often-used method from the second field is the modal method (MM). The application of the MM requires data and information on natural modes, which makes it more complicated for application while yielding results with similar accuracy compared to Ko's displacement theory (Reference Esposito and GherloneEsposito and Gherlone, 2020). In the third group, especially physics-informed neural networks gained interest during the last years, but in studies like in (Reference Xu, Cao, Yuan and MeschkeXu et al., 2023), there is no information given on the required sensor quantity. Besides, there is a lot of effort needed for implementation and training, e.g. due to preparation of training data (Reference Go, Noh and Hyuk LimGo et al., 2025). The inverse finite element method (iFEM) is mainly referred to in literature for the fourth field of shape sensing techniques. It is, like the other techniques from the second and third group, complicated to implement due to specific, specialised element formulations and requires a lot of sensors to yield high accuracy displacement predictions (Reference Esposito and GherloneEsposito and Gherlone, 2020). Therefore, iFEM is only sensitive to implement if fibre optical sensors (FOS) are available, which are more complicated to install and to read data then from, e.g., simple strain gauges (Reference Roy, Esposito, Surace, Gherlone, Tessler, in Rizzo and MilazzoRoy et al., 2023). For the purpose of this paper, Ko's displacement theory will be used for shape sensing in the following. The choice is made due to the simple implementation for data processing in a first WSN concept for shape and load sensing and the ability to deal with only few simple sensors, like strain gauges, for accurate results, as demonstrated in the study of (Reference Esposito and GherloneEsposito and Gherlone, 2020).
A WSN is a form of a sensor network where one or more sensors are combined into a so-called sensor node. Sensor nodes therefore consist of the sensors, in case it is needed analogue-digital converters and amplifiers, a microcontroller, a power supply, and a transmission interface (Meyer zu Westerhausen et al., Reference Meyer zu Westerhausen, Raveendran, Lauth, Meyer, Rosemann, Wawer, Stauß, Wurst and Lachmayer2024b). In the literature, two forms of WSNs could be found, distinguished by their form of data transmission: infrastructured or non-infrastructured networks. Figure 1 shows a graphical comparison.
Infrastructured networks are based on an existing network, like e.g. a wireless local area network (WLAN), in which the sensor nodes join in (see Figure 1 (a)). Therefore, all the sensor nodes communicate with an access point. In contrast, non-infrastructured, and especially ad-hoc networks in this type of WSN as shown in Figure 1 (b), are self-establishing, organising and adapting networks in which each participant can communicate directly with all other participants within range (Reference Patel, Kumar and in DwivediPatel and Kumar, 2018). From these characteristics results a great importance of transmission protocols, which require a good knowledge of the network, the used hardware, and the data processing (Reference Giancol, Martello, Cuomo and Di BenedettoGiancol et al., 2005). Since the scope of this paper is only a first prototype and the use of ad-hoc networks is particularly useful in larger environments than laboratories, only the infrastructure-based WSNs will be considered in the following. Therefore, the term WSN will be used in the following with referring to infrastructure-based WSNs.
Figure 1. Sketch of (a) an infrastructured network with an access point and (b) a non-infrastructured network with data transmission between participants (based on (Reference Giancol, Martello, Cuomo and Di BenedettoGiancol et al., 2005))
Applications of WSNs, especially such using strain measurements for monitoring, are found in various fields in literature. Especially for civil infrastructures, WSNs are used for monitoring. For example, (Reference Liu, Shi, Gu, Zhang, He, Wu and WeiLiu et al., 2021) use a WSN with different fibre bragg gratings (FBG) as FOS for monitoring strata deformation in a borehole and (Reference Taher, Li, Jeong, Laflamme, Jo, Bennett, Collins and DowneyTaher et al., 2022) use a WSN with self-developed strain sensors for SHM of fatigue cracks in steel bridge structures. In the work of (Reference Herrasti, Val, Gabilondo, Berganzo, Arriola and MartínezHerrasti et al., 2016), a sensor node is described for wireless monitoring of accelerations, temperature, and strains on the example of a wind energy turbine. The sensor node is compared to a wired system and was found to work well for vibration mode detection and monitoring of environmental conditions. Besides, the WSN measurement's accuracy was comparable to the laboratory's measurement system while being cheaper than it. In the study of (Reference Tantele, Votsis, Onoufriou, Milis and KareklasTantele et al., 2016), a WSN is presented to assess the condition of a highway bridge. For this purpose, three sensor nodes consisting of a strain gauge, a temperature sensor, and a wind sensor are connected to a WSN using WiFi in field testing. The measurements are analysed in different approaches and are displayed with a graphical user interface (GUI). Besides civil infrastructure applications, WSNs are also applied in the field of aerospace. For example, in the work of (Wu et al., Reference Wu, Yuan, Shang and Wang2009a), an aeroplane component was chosen for strain monitoring with a WSN. For this purpose, different sensor nodes are designed and evaluated on the example of a carbon-fibre reinforced plastics (CFRP) wing box as a stiffening element of an aeroplane wing, focusing on a reliable data transmission and routing algorithm. In the work of (Wu et al., Reference Wu, Yuan, Zhou, Ji, Wang and Wang2009b), the sensor nodes of a WSN were used to measure the strains of a scaled aeroplane wing in a representative strength test. However, these sensor nodes are large due to large data transmission units and large shock-proof boxes.
In a comparison of the related works on WSNs for strain measurements it becomes clear that there is a lack of studies on WSNs for shape sensing. Besides, the WSNs and sensor nodes in the presented works are designed without using a methodical approach for product development, like, for example, the VDI 2206 for the development of cyber-physical and mechatronic systems. Furthermore, publications using shape sensing techniques with sensor networks in laboratory environments do not use wireless data transmission, which might be crucial in real-world lightweight applications on large-scale structural components. Therefore, a methodical conceptualisation and functional verification of WSN for shape sensing is presented in the following to address this gap.
3. Design of sensor nodes and a wireless sensor network for shape sensing
From the comparison of related works, it becomes clear that there is a gap in literature in the field of systematically developed WSNs for shape and load sensing. Therefore, a concept for such a WSN is presented in the following, which is developed using the V-model for mechatronic- and cyber-physical systems in accordance to VDI 2206 or respectively, the adapted form as “ 3-level procedure model” of (Reference BenderBender, 2005) with a focus on system, subsystem and component levels. The essence of this model is the decomposition of the system into its elements on the left part of the “V” and the stepwise integration of elements and subsystems into the system along the right part of it. In between of these two parts, a continuous verification and validation of the system under consideration is carried out (Reference Verein Deutscher Ingenieure, Verein Deutscher Ingenieure and Düsseldorf.Verein Deutscher Ingenieure e.V., 2021). Besides, it would be possible to use other models for development processes as well, e.g. “The Mobius Strip Model of IoT Development Process” (Reference Lee, Cooper, Hands and CoultonLee et al., 2022). This model is more specific for the IoT development itself, but focusses less on the implementation for a mechatronic product. Therefore, the V-model is considered further, especially due to the more detailed focus on different system levels. A sketch of the V-model with the levels and their meaning in the context of the WSN development is shown in Figure 2.
Initially, requirements for developing the WSN are defined on the system level. In the case of this paper, the WSN should be applied for shape and load sensing of an aeroplane wing box, which leads to requirements on the sensor choice as well as the size of the sensor nodes. Concerning the last point, the first requirement is to design the sensor node as small as possible with a minimal height so it fits between the wing box and the aerodynamic profile of the wing. Besides, for the application of an aeroplane component, a lightweight design is also essential, so the sensor nodes should be of minimal weight. For the purpose of shape and load sensing, the use of strain sensors is required, and the data acquisition and analysis should be designed for static and dynamic applications with an adaptive measurement frequency as well. Furthermore, the WSN should be designed as an infrastructured network to reduce the risk of data loss due to complicated and maybe incorrect routing of data transmissions. Therefore, a central access point is required for the network. These few requirements are only an excerpt from a catalogue of requirements for the WSN, which is used for further development.
Figure 2. Tailored V-model for the development of wireless sensor networks
From the requirements, a first draft of the system design is derived with the functional units of a WSN. Based on the identified functions, a morphological box is created with suitable solutions for the functions, which must be fulfilled by the sensor nodes on the subsystem level in the hardware analysis and conceptualisation. This method was chosen because of its suitability to find various characteristics for parameters of a considered structure and it was already successfully applied in IoT applications in (Reference May, Glatter, Arnold, Pfeffer and LanzaMay et al., 2024). From this morphological box, three concepts are chosen for possible designs of a sensor node as a subsystem of the WSN, which is shown in the excerpt of the morphological box in Figure 3. These solutions are evaluated regarding their suitability for the defined requirements using the scoring method as described by (Reference Schlattmann and SeibelSchlattmann and Seibel, 2024), where the fulfilment of the requirements is used as criteria with different weights. As a result, a concept of a sensor node with strain gauges for measurement is chosen. This choice was made to measure the strains only at a few discrete positions, which would not be able with fibre-based measurement systems since they need to be applied as whole fibres. Besides, the data processing is much easier with simple equipment for strain gauges than for FOS or FBG. For the measurement equipment, a solution with a device combining an A/D converter for signal conversion and an amplifier. In the following, the HX711 will be considered for this, since it is a well-known chip for signal conversion and amplification and because it is not of great cost. It would also be possible to dispense with an A/D converter and use the analogue inputs of a microprocessor, but in this case, a significantly higher number of microprocessors would have to be used, which is not sensible due to an increase in costs for microprocessors. Regarding this data processing, a solution with both, microprocessors and a server is chosen, since the microprocessors could be used for preprocessing the data as well as organising the transmission in specific, readable formats, and the server is used for the shape sensing processing of the whole datasets as well as the visualisation. For data transmission, especially WiFi and Bluetooth are used in infrastructure-based WSNs. Since this paper focuses on using an infrastructure-based WSN, the option of an ad-hoc network is not considered in the solutions in Figure 3. To ensure that data can be collected from as many sensor nodes as possible, Bluetooth is not a suitable solution because it is not developed in a standardised form for the communication of many clients in a network.
Figure 3. Excerpt of the morphological box for sensor node conceptualisation for the WSN
Therefore, Solution 1 with a sensor node concept consisting of strain gauges, an HX711 A/D converter and amplifier, a microprocessor and a server for data processing and WiFi for data transmission is chosen. Regarding this solution, the software on the subsystem level has to be conceptualised in different modules. One module has to be developed to read the data on the HX711 and convert the digital inputs of a microprocessor into the form of strains. Besides, another module has to be developed for the communication of the microprocessor and the server in WiFi-based network. Lastly, a third module is required for the set-up of a server, which has to include the communication with the sensor nodes and the processing of the received strain data in the context of shape sensing. On the component level, it is chosen to buy hardware components instead of designing them because of low costs for already available hardware. This has the advantage of building up on already existing code for the software on the component level. Since microprocessors from Arduino have this advantage due to a large community, a microprocessor from Arduino is chosen over other solutions, like e.g., Raspberry Pi. Furthermore, the use of a Raspberry Pi Pico W was evaluated in a preliminary study, but the availability of many libraries for Arduino, especially with an interface to the HX711 A/D converter and amplifier, made it easier to implement in a first prototype.
In the product family of Arduino, the Arduino Nano 33 IoT is selected due to the already existing WiFi module for data transmission. This makes it possible to design the sensor nodes for a minimal size while saving costs for further hardware and additional software due to more interfaces. Because of the choice for an HX711 A/D converter and amplifier, software for such could be used as a basis as this component is widely used for load cells, which are based on strain gauge measurements. Therefore, the software module for reading, converting and amplifying the strain data is built on that basis and is changed regarding the type of gathered data. This is because existing code usually processes the measurements based on a full-bridge strain gauge configuration directly in the form of weight applied to the load cell. In the case of the WSN and the application in this paper, it is not sensible to use a full-bridge configuration since only linear strain gauges are needed to measure strains in one direction to apply Ko's displacement theory for shape sensing. Therefore, the software has to be designed to read the signals and convert them directly to a single strain value instead of further processing it to weights. Therefore, regarding the hardware choice, a quarter-bridge addition is needed for the linear strain sensors. For this purpose, each strain gauge is combined with additional electrical resistances to a Wheatstone bridge. In this work, the choice was made for strain gauges and additional resistances of 120 Ω because such strain gauges are less sensitive to deviations of the insulation resistance (Reference Hottinger and KjaerHottinger Brüel & Kjaer GmbH, 2024). As a result of the conceptualisation and design, a prototype of a sensor node is resulting as shown in Figure 4. In the depicted prototype, only two strain gauges with quarter-bridge completions and HX711 are used. Due to the number of digital inputs of the Arduino Nano 33 IoT, it is possible to install up to eight of these sensor measurement systems. However, this is not always sensible due to the need for longer cables from the HX711 to the microprocessor, which leads to decreasing signal quality and makes the measurement system less robust. Therefore, the prototype is designed with short cables because this should also be the case in later installations.
Figure 4. Prototype of the developed sensor node for strain measurements for shape sensing
The software developed for strain measurements is installed and tested on the prototype. Since it was found that the timestamp on the acquired data drifts, the software is redesigned and updated so it gets the actual time as a response from the server after each data transmission. This allows to keep the influence of the drift in the time minimal. The server itself is designed as a Python flask server, which can be accessed from a browser, where the data of the sensor node is visualised. After these tests on the component and subsystem level are finished, a functionality validation is required on the system level following the V-model in Figure 2. For this purpose, the case study in the following Section 4 is conducted.
4. Case study for functionality validation on a test rig
To show the applicability of the developed and presented concept of a WSN for shape sensing, it is tested on a demonstration part on a test rig in the SCALE research building of the Leibniz University Hannover. For this purpose, an aluminium rectangular tube made of EN AW 6060 with a length of 2000 mm, 120 mm x 60 mm cross-section, and 4 mm thickness is used, which is clamped at one end and is loaded by a servo-hydraulic cylinder at the free end. The load is applied with 900 N in the middle of the part's width, so only bending with elastic deformations is considered as a load case, as shown in Figure 5.
A rectangular tube is chosen to demonstrate a simplified part of a wing box, which is used in aircraft wings for stiffening the aerodynamic profile. Accordingly, the wing box is mainly subjected to bending by buoyancy forces, which is built up in the described load case for the aluminium rectangular tube. For shape sensing with a WSN, a total of ten strain gauges are applied on the top surface of the part, which allows to collect load data in the sense of a gentelligent component. All sensors have the same distance from one another along the part's length and are positioned symmetrically to the length axis on two parallel lines for the calculation of deflections of the two lines as well as the distortion between them. For more details on the calculation, see (Reference Ko, Richards and van TranKo et al., 2007).
Figure 5. Load case on the SCALE test rig with an aluminium rectangular tube as demonstration part
A sketch of the sensor positions for the case study is shown in Figure 6. Besides, the positions of the three microprocessors combining these sensors to sensor nodes are marked in the sketch, and each sensor is allocated to the corresponding microprocessor. The positions of the sensor nodes are chosen in a way that the distance to sensors with a lower signal-to-noise ratio (SNR) is shorter than to the others. This is selected in accordance with the methodology presented in (Meyer zu Westerhausen et al., Reference Meyer zu Westerhausen, Kyriazis, Hühne and Lachmayer2024a) to keep the influence of longer cables on the signal quality of sensors with already less good SNR as low as possible. The data collected by the sensor nodes is processed as described in Section 3. Each sensor node transmits its data to the server in the form of a *.json file with the sensor ID, the sensor node ID, the strains during the measurement period as well as the timestamps. In this case study, the measurements are performed with a frequency of 10 Hz as an exemplary case but could also be carried out with up to 80 Hz with the HX711. Each measurement period between the start and the data transmission is 3 s to make the WSN work in real-time. To successfully transmit the data to the server, the sensor nodes must be accepted as clients. Therefore, each sent data package is equipped with the server's SSL certificate key, which must be accepted to ensure trustworthy data.
Figure 6. Sensor positions on the top surface of the rectangular tube for the case study
After each sensor node's data is received at the server after the measurement periods, it is plotted on the basis of the server's URL website in the form of a strain-over-time curve. Besides, the strains are sampled and used to calculate the displacements with Ko's displacement theory. This calculation is performed in a Python script, and the results are further processed to reconstruct the whole deformation field of the part using Python's radial bias function (RBF). The reconstructed displacement field with the deformed shape is shown in Figure 7 for one of the measurements timesteps of as a scatterplot. For each timestep, such a scatterplot is created which allows to scroll through the deformation history on the server URL website. This allows to have the strain-over-time curves on the one and the visualisation of the deformation field on the other hand, which makes it possible to get detailed insights into the load history of the gentelligent component. Using this WSN and the data processing on the server therefore enables the further development of techniques for design adaptions based on acquired information. Besides, it enables the monitoring of products and their structural components in industrial applications, like e.g. in the automotive and aeroplane industry.
Figure 7. Exemplary plot of a reconstructed deformation field from strain measurements in the case study
Therefore, these results validate the functionality in the step of the system integration of the developed sensor nodes as hardware and the corresponding software for data acquisition and transmission. In the next step, the WSN prototype can get developed further as a continuous process to bring it to the next maturity level in accordance to VDI 2206 (Reference Verein Deutscher Ingenieure, Verein Deutscher Ingenieure and Düsseldorf.Verein Deutscher Ingenieure e.V., 2021).
5. Conclusion and future work
The use of wireless sensor networks (WSNs) in the context of sensor-integrating, gentelligent components is of interest from a product development and design point of view to get real-time insights into operational loads of large-scale components for optimisations of future product generations. So-called shape and load sensing techniques could be applied to get a full-field reconstruction of deformations and loads from measurements at discrete positions. In literature, there are a lot of publications on different shape sensing techniques and different applications of WSNs in the context of structural health monitoring (SHM), but the use of a WSN for shape sensing is not emphasised enough. Therefore, a conceptualisation of a WSN for this purpose along the V-model for developing mechatronic and cyber-physical systems is described. To validate the functionality and fulfilment of criteria on the WSN as a system, a case study is carried out on the example of an aluminium rectangular tube under bending load on a test rig at the SCALE research building. In the case study, the functionality of the WSN was demonstrated with three sensor nodes with strain gauges and Arduino Nano 33 IoT microprocessors in an infrastructure-based WSN, where gathered strain data is processed on a server and visualised on a URL website.
However, some issues could still be found regarding the use of the WSN, which have to be optimised in future works. First, the measurement was only carried out with a frequency of 10 Hz. In future works, it should be enhanced to 80 Hz, which is available with the HX711 A/D converter and amplifier, to allow the application in more dynamic tests. Furthermore, the positions of the sensors, as well as the microprocessors, were chosen without the performance of an optimal sensor and sensor node placement. In future works, this has to be addressed to gain the most accurate results regarding the deformation field reconstruction. Besides, various interfaces between different program scripts are needed for data acquisition and processing, which has to be simplified for future applications. Last, the periods between the start of each measurement and the data transmission are chosen with 3 s due to the time needed to process the data. Therefore, in future applications, an optimisation of the program scripts will be carried out to allow shorter measurement periods. This leads to the potential of more real-time applications, which are of interest when a further use for, e.g., a digital twin is considered. Besides these technical aspects, the data management along the workflow of the data acquisition in accordance to (Reference Wawer, Wurst and LachmayerWawer et al.) should be applied, which should yield higher consistency in contrast to only storing WSN acquired data in the end of each monitoring task.
Acknowledgement
The research building SCALE - Scalable Production Systems of the Future and the testing equipment multi-axis, dynamic load test rig were funded by the Federal Ministry of Education and Research (BMBF) and zukunft.niedersachsen, a funding program of the Ministry for Science and Culture of Lower Saxony (MWK) and the Volkswagen Foundation.
Open access
