Sean Flanagan

Abstract

Consistency is a major factor in the performance of table tennis players. The integration of sensors into athletic equipment has been shown to elevate athletic performance and consistency. The goal of this project was to integrate a system of sensors into a table tennis racket and reliably transmit data to a mobile app for the purposes of game analysis and coaching. A shot classification algorithm was developed using data from a gyroscope placed inside the handle of the racket connected to an Arduino Pro Mini. Also, a 4 x 4 grid of piezoelectric sensors was placed under the face of the racket to detect the impact location of each shot. The data for each shot were transmitted via Bluetooth to an Android Tablet for real-time viewing. The prototype was able to able to identify, classify, and relay location data over 90% of the time during a session. These results suggest that the device will provide consistent and accurate feedback that could be immediately implemented to aid table tennis players of all ages and skill levels in the development of their game.

Introduction

To compete on a high level, professional table tennis players must have rigorous physical and mental training. Table tennis athletes only have 0.2 to 0.4 seconds to react to an approaching ball (Kondrič, Zagatto, & Sekulić, 2013), and after a long match, players can begin to tire. Table tennis players must train and condition their bodies for the physiological demands of the game to maintain their peak performance (Kondrič, Zagatto, & Sekulić, 2013).
Tennis, which is very similar to table tennis, has many sensors developed for user improvement. Zepp, Babolat, and Sony are three companies that have all successfully developed marketable tennis sensors that record data relevant to a user’s gameplay and provide suggestions for improvement. Rafael Nadal, a prominent tennis player, uses a Babolat smart racket that allows his coach to make any necessary tweaks to his training plans. Nadal and his coach are effectively able to utilize the data provided by his racket and improve the pro tennis player’s performance. This technology is very useful to professionals and novice tennis players. However, there is currently no app-based feedback system available for the sport of table tennis (Tracy 2016).
 Since there is currently no feedback systems for the sport of table tennis, this project focuses on initial integration of sensors and basic app development. Additionally, the project seeks to provide information on possible improvements and future implications of smart table tennis systems.

Methods

Setup
A 6 degrees of freedom (DOF) inertial measurement unit (IMU) MPU-6050 GY-521 Breakout Board Accelerometer/Gyroscope used and a program obtained from www.i2cdevlib.com was then used to calibrate the MPU-6050. Using this program offsets were calculated for each raw value that the MPU-6050 output (X Acceleration, Y Acceleration, Z Acceleration, X Gyroscope, Y Gyroscope, and Z Gyroscope). An inclinometer app on a Nexus 7 Android tablet was used to create a flat surface (required by the program) on which the accelerometer could rest and the program could be run. The calculated offsets were then added to the program obtained from the library developed by Jeff Rowberg for improved accuracy.
Shot-Classification
A simple shot classification algorithm was then developed to categorize shots based off of the yaw, pitch, and roll values obtained from the MPU-6050. After a slight change to the MotionApps C++ header file in the MPU-6050 Library, the MPU-6050 recorded the values for yaw, pitch, and roll on a -90 to 90 scale instead of the default 0 to 180. The MPU-6050 was then wired to an Arduino Uno and placed on the handle of the racket. The racket was then tilted in the backhand orientation and the yaw and pitch values were recorded to obtain viable ranges for both backhand shots as well as individual topspin and backspin shots. The same procedure was then followed to obtain values for the forehand orientation. The MPU-6050 example sketch was then adapted using a series of “if else” statements that would subsequently print out the orientation. The sketch was then tested by rotating the racket into all four orientations.
Shot Detection
The ground wires of 16 piezoelectric sensors were connected in series and the data wired were then connected to a Sparkfun CD74HC4067 multiplexer. An Arduino Pro Mini and an HC-05 Bluetooth module were connected to both the accelerometer and the piezoelectric array to enable Bluetooth connectivity and a power supply. A coordinate grid was overlaid onto the 4 x 4 array of piezoelectric sensors with coordinates assigned to each single, pair, and group of four sensors. A series of “if else” statements were then added to the program with the quadruples taking precedence and being checked first in order to avoid multiple readings of the same impact.
Final Prototype
A Butterfly Timo Ball A500 shakehand table tennis racket was acquired from Amazon.com and the red rubber siding of the racket was removed using a combination of sanding and prying with a blade. Then the face of the racket was coated in carpenter’s glue and the 4 x 4 array was placed on top. The glue was left for fifteen minutes to dry. After the racket was sufficiently dry, a Yinhe Mercury II Medium rubber siding acquired from Amazon.com was glued to the face of the racket about 89 N of force were applied to allow the rubber to dry and stick. After 6 hours, the force was removed. The voltage readings for each sensor was recorded over a 15 second time period and the maximum value was used as an offset for each sensor in the multiplexer sketch.
Testing
The HC-05 Bluetooth module was then connected to the Blue2Serial App provided by MacroYau on Github.com. A nine volt battery was connected to the Arduino Pro Mini and the breadboard setup was placed on the back of the racket for testing purposes. The HC-05 was then connected to the Blue2Serial app. Serves were completed 30 times in each of the four orientations with the identification, correct classification, and transmission of location data being relayed and recorded in Microsoft Excel.

Results and Analysis

Identification

    From the 120 trials conducted, 30 separate Backhand Backspin shots, 27 separate Backhand Topspin shots, 28 separate Forehand Backspin shots, and 27 separate Forehand Topspin shots were identified by the system and relayed to the mobile device. 27 of the Backhand Topspin shots were correctly classified out of the 30 total Backhand Topspin shots identified. Similarly, 30 of the Backhand Backspin shots were correctly classified out of the 30 total Backhand Backspin shots identified. 27 of the Forehand Topspin shots were correctly classified out of the 30 total Forehand Topspin shots identified and 28 number of the Forehand Backspin shots were correctly classified out of the 30 total Forehand Backspin shots identified.
    The impact location of each shot was also relayed 112 times out of the 120 total number of shots identified. The number of shots identified out of the total number of shots produced a success rate for accurate identification of 93% . Out of the 120 total trials for the 4 different possible shot classifications, the sensor system correctly classified and relayed shots producing an average 93% success rate for the classification of shots overall. The number of shots where location data was successfully transferred out of the total number of shots produced a success rate for accurate location transfer of 93% .

Fisher’s Exact Testing

A Fisher’s Exact test, typically used with categorical data, was performed on each of the data sets (identification, classification, and location transmission) to determine whether or not the deviation from the null hypothesis was significant.
H0: The tested prototype should identify shots, classify shots, and transfer location data equally as well in each of the four possible orientations.
The Fisher’s Exact Test used to analyze the identification results for all backhand shots yielded a statistical value of 0.118644 which is greater than 0.1 and therefore not significant. Also, the Fisher’s Exact test used to analyze the results of the forehand shots yielded the same statistical value of 0.118644 and again was not significant. The Fisher’s Exact Test used to analyze the classification results for all backhand shots yielded a statistical value of 0.118644 which is greater than 0.1 and therefore not significant. Again, the Fisher’s Exact test used to analyze the results of the forehand shots yielded a statistical value of 0.676628 and again was not significant. The Fisher’s Exact Test used to analyze the location transfer results for all backhand shots yielded a statistical value of 0.118644 which is greater than 0.1 and therefore not significant. Also, the Fisher’s Exact test used to analyze the results of the forehand shots yielded a statistical value of 0.676628 and again was not significant. Since all of the statistical values obtained from the Fisher’s Exact tests were above 0.1, the deviation from the null hypothesis was not significant and the null hypothesis was accepted. All of the shot orientations were considered to identify, classify, and transfer location data equally as well.

Conclusions

The goal of engineering a system of sensors that could fit into the handle of a racket and reliably transmit data to a user or coach was achieved. The sensors all are of a small size where they could easily be combined into a printed circuit board or some other smaller device and fit into the handle of the racket. Battery power could also be optimized using a lithium ion battery and fit into the handle. The software used in this project would not be compromised and still would function on a high level and provide accurate feedback to a user. With a success rate in all orientations over 93%, this system could easily be fine-tuned and presented commercially for all table tennis players. The prototype could be marginally improved. The hardware and breadboard on the back of the racket could also be combined into a printed circuit board and inserted into the handle of the racket to allow a player to use both sides of the racket. A second piezoelectric array could be added to the other side of the racket to create a fully functional racket and allow players to use both forehand and backhand without using the same side for both orientations. The app could also be improved. Currently, the app only receives and displays the data from the sensors. This could be improved to track the data and use in-app analysis to give the user more information about their skills and improvement. The app could display graphs of target angles, locations, and shot numbers to compare against the user’s data. Overall, both the app and the racket could be improved to reflect more of a player’s game.