Flexible arch-shaped triboelectric sensor based on 3D printing for badminton movement monitoring and intelligent recognition of technical movements

With the continuous advancement of sensing technology, big data, and artificial intelligence, intelligent data monitoring and analysis systems are increasingly permeating various domains.^1–3 Sports monitoring aligns with this trend toward intelligence, where the utilization of systematic and scientific sensing systems to monitor sports indicators has become a pivotal concern in modern sports.^4–6 Among popular sports disciplines, badminton is known for its intricate and variable technical movements as well as its renowned dynamic speed and precision. Continuous acquisition, identification, and analysis of data can be employed for timely correction of movements and postures while updating training plans, which greatly contributes to enhancing badminton performance.

In traditional methods of badminton technique research, video analysis is the main method. This method mainly utilizes machine vision technology to analyze the postures, footwork, and technical movements of badminton players.^7,8 However, video analysis contains some shortcomings, such as being affected by the shooting angle, high requirements for the site, and privacy disclosure, which is not suitable for long-term tracking and monitoring of athletes. With the rapid advancement of wearable sensor technology, an increasing number of researchers are utilizing accelerometers, gyroscopes, inertial measurement units (IMUs), electromyography measuring instruments, and other sensors to monitor athletes’ training loads, arm swing movements, and muscle strength analysis.^9–11 However, the widespread implementation of distributed body area networks has brought to light certain issues with these rigid sensors, such as discomfort from long-term wear, significant interference for athletes, and high power consumption within the system.¹² Due to the inherent characteristics of triboelectric nanogenerator (TENG), such as no external power supply, simplistic structure, extensive material options, and adaptability for flexible device fabrication, it has found widespread applications in motion posture sensing and movement recognition,^13–19 showing great potential in the field of intelligent sports.

Currently, imperfections in the manufacturing process have become one of the main obstacles to the further development of TENG, impeding its industrialization.²⁰ The advancement of 3D printing technology streamlines the design and fabrication process of flexible sensors, offering a potential avenue for the large-scale production of triboelectric sensors with distinctive structures.²¹ The Fused Deposition Modeling (FDM) technology, which is a low-cost and fast-forming technique in 3D printing, has been successfully commercialized.²² Using thermoplastic elastomers and shape memory polymers with good mechanical properties as raw materials makes it easy to prepare various spatial structures of flexible TENGs,^23–25 which have advantages over traditional methods for manufacturing flexible devices. Flexible sensors based on FDM technology that have been reported so far exhibit a certain level of skin compliance.^26–29 However, the skin undergoes not only bending deformation but also a certain degree of stretching deformation when the moving parts of the human body, such as the elbow joint and wrist joint, bend, and twist within a wide range. At this moment, the thin-film flexible sensors that can only bend and fold will impose certain restrictions on joint movements. Since wearing comfort is crucial for professional athletes,³⁰ discomfort during major deformations might interfere with the completion of athletic techniques. Therefore, there is an urgent need for 3D printed triboelectric sensors that do not restrict human joint skin bending, twisting, and stretching deformations.

In addition, it is noteworthy that reports on the application of triboelectric sensors in badminton are still scarce. Ning et al.³¹ fabricated a triboelectric fiber composed of a liquid metal core and a polyurethane sheath, which can form a large-area TENG through weaving. When attached to the forearm, it can be utilized for monitoring the striking movements in badminton. Wang et al.³² developed a stretchable sensor based on ultra-tough and high-elastic electrospinning film, which is pasted on the arm and applied to the posture monitoring of forehand, backhand, spike posture, and other positions. However, in these studies, the visual difference of the voltage signal waveform collected by the oscilloscope and other instruments is used to distinguish different badminton technical movements. The method of distinguishing technical action in this way is subjective, time-consuming, and prone to inaccuracies, with limitations in deeply mining technical action information. Inspired by the novel integration of TENG and machine learning in the domains of human–computer interaction interface,^33,34 health monitoring,^35,36 and intelligent sports,³⁷ incorporating machine learning into the analysis of badminton sports data collected from multiple triboelectric sensors can provide a more comprehensive evaluation of athletes’ performance. Therefore, in order to facilitate the widespread implementation of triboelectric sensors in badminton sports monitoring, it is imperative to integrate machine learning techniques for comprehensive analysis of badminton sports data and enhance data utilization.

To address the discomfort associated with wearing the current 3D-printed flexible triboelectric sensor during extensive joint movements and the lack of in-depth triboelectric sensor data analysis in the field of badminton, a 3D-printed flexible arch triboelectric sensor (FA-Sensor) is proposed in this paper. On this basis, an intelligent multi-sensor badminton monitoring system is constructed based on the FA-Sensor, the neural network algorithm, and the master computer. The FA-Sensor is characterized by its lightweight design, affordability, and rapid customization capabilities. The elastic TPE housing and arched shape enable it to minimize interference with human motion during bending (10°–150°) and twisting (20°–100°) across a wide range. For various articulations of the human body (such as the wrist, elbow, shoulder, knuckle, knee, waist, neck, etc.), it can be quickly customized to accommodate different sizes of sensors while ensuring a stable and consistent response to the movements across multiple articulations. In addition, the badminton sports monitoring system enables online monitoring, real-time feedback of the multi-joint motion information, and accurate identification of four typical badminton movements. The system enables athletes to monitor, analyze, and enhance badminton technology while also exhibiting promising potential for application in other intelligent sports domains.

The FA-Sensor is capable of detecting the motion of human joints. When utilized for badminton monitoring, the FA-Sensor can be attached to multiple joints on the human body in order to achieve detailed action monitoring and technical movements’ pattern recognition through signal collection, transmission, and processing. The FA-Sensor has a simple structure consisting of a dual-arched shell, a layer of silicone rubber, and a conductive fabric layer, as shown in Fig. 1(a). Among them, the double arch shell of TPE consists of two flexible arches, one nested inside the other. The material’s excellent mechanical flexibility, combined with the well-designed shape of the double arch, imparts impressive bending performance and resilience to the shell. Conductive fabric layers are adhered to the inner walls of the upper and lower sides of the dual-arched shell. The upper side of the conductive fabric serves as an electrode, while its other side is coated with a silicone rubber film as a negative triboelectric layer. The lower conductive fabric functions as both a positive triboelectric layer and an electrode.

The fabrication process of the FA-Sensor is shown in Fig. 1(b). First, the double arched shell is fabricated using FDM 3D printing technology. The three-dimensional structure of the dual arch is custom-designed using modeling software (SolidWorks 2024). The thermoplastic polyurethane elastomer (TPE-83A, Esun) is utilized as the material for the double arched shell. The TPE-83A from Esun comes in two different colors. Other than the color, the performance is identical and will not affect the test results. In addition, the dual-arched shell is integrally fabricated using an FDM 3D printer (P1S, Bambu Lab). Second, a silicone rubber film is fabricated using Ecoflex silicone products (Ecoflex 00-30, Smooth-On, Inc.). Ecoflex 00-30 rubbers are mixed 1A:1B by weight. Transfer the mixture into a rectangular mold (35 × 9.5 × 1 mm³) and allow it to stand at room temperature for a duration of 4 h prior to curing. Third, commercial conductive tape [0.1 mm plain double-sided conductive tape, Ubixuan Tape (Hangzhou) Co., Ltd.] is used as conductive fabric and cut into a rectangle (35 × 9.5 mm²). Finally, the materials are assembled, and wires are drawn from the conductive fabric on both the upper and lower sides to obtain the FA-Sensor, as shown in Fig. 1(c). It can be seen that the FA-Sensor is easy to bend and twist. Due to its arched shape, commendable flexibility, and minimal mass (2.5 g), the FA-Sensor has little interference with badminton when it is attached to human joints.

The test system consists of an electrometer (6514, Keithley), a vibration isolation optical platform (250 × 250 mm², HENGYU), a stepping motor (YH57BYGH56, CNYOHO), a stepping motor controller (YF-27, YEYY), an embedded system based on a microcontroller (STC8H1K28-36I-LQFP32, STC), and structural components manufactured by a 3D printer (P1s, Bambu Lab) using PLA material (PLA+, eSUN), as shown in Fig. S1. When the bending test is performed, the FA-Sensor is affixed to the test machine using tape, and the structural components undergo repetitive bending at various angles and frequencies of motion controlled by a stepper motor controller. When the torsion test is performed, the two ends of the FA-Sensor are fixed to the rotating support and the stationary support, respectively. When the rotating support rotates with the stepper motor, one end of the FA-Sensor remains stationary while the other end rotates. The torsion angle is the rotation angle of the motor. The output signal from the FA-Sensor is collected and recorded by the electrometer. It should be noted that when recognizing the movement patterns of badminton, an oscilloscope (MSO2024B, Tektronix) and a multi-channel data acquisition card (USB-6009, NI) were used due to the need for multiple signal acquisitions.

The FA-Sensor belongs to the contact–separate mode TENG. In this case, the silicone rubber film serves as a negative triboelectric layer due to its excellent electron affinity and ease of acquiring electrons. Compared to silicone rubber, the conductive fabric tends to lose electrons more easily, serving as a positive triboelectric layer. During the movement of human joints, there is continuous contact and separation between the silicone rubber film and the conductive fabric. The charge transfer process within a single cycle is shown in Fig. 2(a). In the initial state, the two triboelectric layers are separated (original state), appearing electrically neutral with no current output. When the FA-Sensor is fully bent during joint movement (state I), its two triboelectric layers come into contact, leading to the redistribution of surface charges due to triboelectric charging and electrostatic effects, resulting in the formation of equal and oppositely induced charges on the contact surface. During state II, as FA-Sensor gradually stretches with joint movement, the separation between the two triboelectric layers increases, resulting in a weakened attraction of the silicone rubber film to the surface charge of the conductive cloth on the lower side. Consequently, under the influence of a potential difference between two electrodes, electrons are transferred from top to bottom through an external circuit, thereby generating an instantaneous bottom-up current externally. After the complete release of the FA-Sensor (state III), the maximum distance between the two triboelectric layers is achieved, facilitating electron transfer through the external circuit to establish a temporary state of charge equilibrium, which manifests as neutrality. During state IV, as the FA-Sensor gradually flexes with joint movement, a portion of electrons is transferred from bottom to bottom through the external circuit due to the attractive force exerted by the silicone rubber film, resulting in an instantaneous top-down current. Through the flexion and extension movements of human joints, the two triboelectric layers of the FA-Sensor continuously engage and disengage, thereby generating a continuous output of alternating current electrical signals.

Using COMSOL Multiphysics 2023 software, the electric potential distribution of the FA-Sensor in an open circuit condition was simulated, as shown in Fig. 2(b). The three simulation diagrams correspond to the states III, II, and I in Fig. 2(a), where different colors indicate the magnitude of the potential at each location. It can be seen that the electric potential difference is the largest when the two triboelectric layers are completely separated. As the FA-Sensor bends, the two layers approach each other, and the potential difference gradually decreases. When the two layers are in full contact, the potential difference reaches its minimum. Conversely, as FA-Sensor expands, the two layers move apart, and the surrounding electric potential gradually increases. This result is consistent with the working principle of the contact–separate mode of TENG.^38,39

The posture of the upper arm directly affects the standardization of badminton technical movements. To verify the effectiveness of the FA-Sensor in monitoring human joint motion, it was tested when affixed to the wrist and elbow, as shown in Figs. 4(a) and 4(d). The results of detecting the wrist bending angle using the FA-Sensor are shown in Fig. 4(b). It can be seen that when the wrist bending angle is 10°, the open-circuit voltage amplitude is 0.345 V. As the wrist bending angle gradually increases, the open-circuit voltage amplitude also increases. When the wrist bending angle reaches 50°, the open-circuit voltage amplitude reaches 0.72 V. As shown in Fig. 4(c), there exists an approximately linear relationship between the wrist bending angle and output voltage amplitude, which aligns with the variation trend within the range of 10°–70° presented in Fig. 3(c). When the elbow bending angle changes from 30° to 150°, the peak open-circuit voltage of the FA-sensor increases from 0.51 to 3.1 V, as shown in Fig. 4(e). The variation trend of voltage amplitude mirrors that displayed in Fig. 3(c), as demonstrated by Fig. 4(f). In addition, owing to its arch shaped design, the FA-Sensor exhibits certain extensibility after being bent while causing minimal interference with joint motion. This indicates that the FA-Sensor effectively monitors both wrist and elbow bending motions in humans.

In badminton sports, some multi-degree-of-freedom joints, such as the shoulder joint, undergo torsion in addition to bending. Therefore, FA-Sensor’s output response in terms of current and voltage under various degrees (20°, 40°, 60°, 80°, and 100°) of torsion was tested, with results shown in Figs. 5(a) and 5(b). It can be observed that the FA-Sensor also produces output responses during torsion. When the torsion angle increases, the open-circuit voltage and short-circuit current of the FA-Sensor gradually increase. Compared to bending motions, the waveforms of voltage and current are sharper, and the amplitudes are larger. This phenomenon can be attributed to the increased force exerted on the FA-Sensor during twisting and its quicker reset upon release. As the degree of torsion increases from small to large, the peak values of open-circuit voltage and short-circuit current gradually increase. This indicates that the magnitude of the voltage amplitude intuitively represents the extent of torsion. Furthermore, some technical movements in badminton require coordination among multiple joints throughout the body. To address the challenge of monitoring joints with varying sizes, sensors of diverse dimensions were fabricated (Fig. S4), capitalizing on the facile and rapid customization offered by FA-Sensor through 3D printing. Scaled versions of FA-Sensors were applied to finger joints, knee joints, the neck, and waist, with test results shown in Fig. 5(a)–5(f). It is evident that FA-Sensors with different sizes can be easily affixed to mobile body parts. At different bending degrees, these FA-Sensors can effectively distinguish the extent of bending using the amplitude of the voltage. This indicates that these FA-sensors are sensitive to bending and twisting, and can be flexibly customized for different mobile body parts, which is suitable for monitoring various mobile body parts of the human body.

Forehand serve, backhand serve, forehand hook under the net, and backhand hook under the net are the basic technical movements in badminton. To monitor movement posture and identify technical movements, three FA-sensors are fixed on the wrist joint, elbow joint, and shoulder joint of a player’s dominant arm, respectively (the sensors corresponding to the three joints are called Sensor-W, Sensor-E, and Sensor-S). Multi-channel oscilloscopes were used to collect the output voltage signals of the three sensors, and the results are shown in Fig. 6. It can be observed that the distinct batting movements of the player correspond to diverse output voltage signals. The number of voltage peaks indicates the number of strokes, and the size of the peaks reflects the bending angle of joints during the stroke. Variations occur in the output voltages among these three sensors when executing specific technical movements, thereby revealing distinct motion states across all involved joints. By combining these output voltage signals from all three sensors, it becomes possible to further analyze players’ postural information during various technical movements.

Taking the forehand as an example, the signals from this technical movement are shown in Figs. 6(a) and 6(b). It was observed that the sensor signals had high repeatability when performing this movement multiple times, easily identifying three instances of striking. When analyzing the sensor signals of a single technical movement individually, they can be correlated with the four stages of forehand serve: stage I: raising the racket, stage II: turning the body and pulling the racket, stage III: power hitting, and stage IV: withdrawing the racket. In stage I, the arm is relaxed in preparation for the stroke, with minimal movement, and all sensors show minimal electrical output. In stage II, the three joints exert force, with particular emphasis on the wrist joint as the primary force generator, exhibiting a significant angle of flexion. Meanwhile, both the shoulder joint and axial joint undergo minimal bending angles. Sensor-W and Sensor-S undergo minor deformations, with small areas of friction within the functional layers of these two sensors, transferring less charge and producing weak electrical signals. At this moment, the wrist tilts backward, and then rapidly and steadily propels forward with a Sensor-E that exhibits significant flexion, thereby generating a distinct electrical signal. During stage III, in order to accurately control the direction of the ball and coordinate the force of the elbow and shoulder, the shoulder joint and elbow joint have different degrees of bending, resulting in significant bending of Sensor-W and Sensor-S. The functional layer within these sensors undergoes sufficient friction, facilitating substantial charge transfer and generating obvious electrical signals. In stage IV, upon completing the shot, prompt arm recovery is essential to prepare for subsequent serves or anticipate the opponent’s return. As the elbow and wrist joints bend, significant bending occurs in Sensor-W and Sensor-S, which output electrical signals, while the wrist joint moves minimally, resulting in no signals from Sensor-E. In addition, as shown in Figs. 6(c)–6(h), it can be seen that technical movements such as backhand serve, forehand hook under net, and backhand hook under net have similar results. By analyzing the signals from the three sensors on the dominant arm, detailed information on the completion and specifics of various badminton techniques can be discerned. This underscores the practical significance of the FA-sensor in facilitating a comprehensive analysis of badminton’s technical movements.

To perform pattern recognition of typical badminton technical movements, a badminton sports monitoring system based on the FA-Sensor was built. The system consists of three FA-Sensors, a multi-channel data acquisition card (USB-6009, NI), and master station software based on NI Labview and MathWorks Matlab. When using this system, the three FA-Sensors need to be affixed, respectively, to the wrist, elbow, and shoulder joints. After synchronously collecting signals from the three sensors using the data acquisition card for signal feature extraction, the master computer performs feature extraction, technical movement pattern recognition, and feedback display of the signals, as shown in Fig. 7(a). In the signal processing process, a smoothing filter (rectangular moving average, half-width moving average parameter is 20) is used to filter the signal. The LabVIEW MathScript module allows for mixed programming with NI LabVIEW and MathWorks Matlab to implement pattern recognition of badminton technical movements based on a shallow multilayer feedforward neural network. Feature extraction is realized based on badminton movement feature analysis. According to the previous analysis, the timing of coordination between each joint and the range of motion for each joint during each technical movement are distinctly different. Therefore, a six-dimensional feature vector is selected as the input, comprising the peak amplitudes of the sensing signals of the three joints, the peak time difference between the shoulder joint and the elbow joint, the peak time difference between the shoulder joint and the wrist joint, and finally, the time difference between the peak and the trough of the shoulder joint, as shown in Fig. 7(b). The P-Find algorithm²⁷ is used for feature extraction. As shown in Fig. 7(c), the feedforward network of the shallow multilayer feedforward neural network used has ten hidden layers composed of sigmoid neurons, and the output layer uses a sigmoid activation function. The number of input features is six, while the output values are four types of technical movements. The training algorithm employed is the quantized conjugate gradient, and the performance evaluation is based on cross entropy error.

A total of 142 sets of observations were collected for pattern recognition, and the results are shown in Fig. 7(d). From the confusion matrix, it can be seen that the accuracies of the training set, validation set, and test set are 97.0%, 100%, and 95.2%, respectively. The overall recognition accuracy was 97.2%, with only two instances of forehand service errors mistakenly identified as backhand service and backhand hooks. The results indicate that the algorithm can effectively recognize these four typical badminton technical movements. Finally, the integrated system was subjected to testing, and the test site is depicted in Fig. 7(e) and Video S1. The master computer’s interface is shown in Fig. 7(f), featuring a settings area and display areas for the sensor signals of the three joints. The test results show that the system can effectively monitor the posture information of the shoulder, elbow, and wrist joints. In addition, using the collected posture information, it can accurately identify four types of typical technical movements: forehand serve, backhand serve, forehand hook under the net, and backhand hook under the net.

In this work, the FA-Sensor is a lightweight (2.5 g), low-cost, and easily rapidly customizable flexible triboelectric sensor for bending and torsion, which can be used in badminton sports monitoring and intelligent technical action recognition. Thanks to the flexible 3D printing technology, its size can be quickly customized according to needs, making it suitable for various human body active parts (wrist joints, elbow joints, shoulder joints, finger joints, knee joints, waist, neck, etc.). The elastic TPE shell and arch shape provide FA-Sensor with minimal interference with human motion during extensive bending (10°–150°) and torsion (20°–100°). Building on this, the effectiveness and reliability of using the FA-Sensor’s output voltage as a sensing parameter for bending and torsion, as well as its adaptability to different human body movement parts, were studied from theoretical and experimental perspectives. Experimental results have shown that the FA-Sensor is responsive to both bending and torsion. The peak value of the FA-Sensor’s output voltage has a defined functional relationship with the device’s bending and is insensitive to the frequency of excitation. When the bending angle is within the range of 10°–70°, the sensor’s sensitivity is 7.98 mV/°. When the bending angle is between 80° and 150°, the peak value of the output voltage rapidly increases, with sensitivity at 29.28 mV/°. After 3200 cycles, FA-Sensor’s output is stable. In addition, for different human body active joints, it can be quickly customized into different sizes of sensors and provides a stable and repeatable response to a variety of human movements. Finally, an intelligent badminton sports monitoring system based on three FA-Sensors, a multi-channel acquisition card, neural network algorithms, and a master computer was built. This system can achieve online monitoring, real-time feedback, and pattern recognition of four typical technical movements (forehand serve, backhand serve, forehand hook, and backhand hook) of the dominant hand’s multi-joint motions for badminton enthusiasts, with a recognition accuracy rate reaching 97.2%. This system can assist coaches and enthusiasts in monitoring, analyzing, and improving badminton techniques and can be extended to other sports for intelligent monitoring, holding great potential in the field of sports intelligence monitoring and analysis in the era of big data.

The supplementary material encompasses Fig. S1: Cyclic (a) bending and (b) torsion test platform; Fig. S2: The relationship between the bending angle and the amplitude of the output voltage; Fig. S3: Output voltage and power of FA-Sensor under different loads; Fig. S4: Different sizes of 3D print-based F-sensors; and Video S1: Badminton sports monitoring system test.

The work was supported by the Fundamental Research Program of Shanxi Province (Grant Nos. 202303021212285, 202303021212284, and 202303021211195), the Key R&D Project of introducing high-level scientific and technological talents in Lvliang City (Grant No. 2022RC09), the Key R&D Project of Lvliang City (Grant No. 2023SHFZ27), the Teaching Reform and Innovation Project in Higher Education Institutions of Shanxi Province (Grant No. J20231342), and the University-Industry Collaborative Education Program (Grant No. 231002608111356).

The authors have no conflicts to disclose.

Y.Y. and L.J. contributed equally to this work.

Yun Yang: Conceptualization (equal); Data curation (equal); Formal analysis (equal); Funding acquisition (lead); Investigation (equal); Supervision (equal); Writing – original draft (lead). Lei Jia: Data curation (equal); Formal analysis (supporting); Investigation (equal); Validation (lead); Writing – original draft (supporting). Ziheng Wang: Data curation (supporting); Formal analysis (supporting); Software (lead). Jie Suo: Data curation (supporting); Software (supporting). Xiaorui Yang: Data curation (supporting); Validation (supporting). Shuping Xue: Data curation (supporting); Investigation (supporting). Yingying Zhang: Data curation (supporting). Hui Li: Data curation (supporting); Funding acquisition (supporting). Tingting Cai: Conceptualization (equal); Supervision (equal); Writing – review & editing (lead).

The data that support the findings of this study are available from the corresponding author upon reasonable request.