Smart Golf Clubs for Player Enhancement
Final Design
The final design solution included a complete mockup of a driver club. The main components of the design are the driver face, driver head, data acquisition system (DAQ), driver handle, driver shaft, receiver enclosure, and receiver display which are explained in more detail below.
Figure 1: Final Design Components
Figure 2: Driver Club Final Design
Driver Face
The final sensing strategy is located on the inside of the acrylic driver face. The design incorporated a multi-layer sensing strategy composed of an array piezoelectric polyvinylidene difluoride (PVDF) thin film transducers. The PVDF sensor strips were used to identify impact location by normalizing the sensor readings based on the readings from impact in the center of the acrylic face. The larger PVDF sensor was calibrated using an impulse hammer (PCB® Modally Tuned® Impulse Hammer, Model: 086C03). The equation obtained from the calibration curve was used to determine the impact force.
Figure 3: Sensor Layout Isometric View Final Design
Figure 4: Sensor Layout Top View Final Design
A detailed description of the sensor fabrication process can be found on the Multimedia page. The sensor layout was encased in and adhered to the back side of the acrylic driver face with insulative epoxy which was then secured to the driver head. The analysis and calibration were conducted by replicating a downsized force based on material properties of acrylic compared to that of titanium-faced driver.
Figure 5: Sensor Layout Adhered to Acrylic with Epoxy Front View
Figure 6: Sensor Layout Adhered to Acrylic with Epoxy Back View
Driver Head
The acrylic driver face was secured to the 3D printed driver head which housed the main DAQ components. The electronic components were encapsulated in a potting mold which was then placed inside the driver head. The encapsulation was accomplished using a bilayer potting method, which consisted of a hard epoxy interior layer to provide rigidity to the printed circuit board (PCB) for deflection prevention and a soft silicone exterior layer to absorb shock. The potting mold featured a tiered design that brought a majority of the mass closer to the bottom rear of the driver head which ensured that the moment of inertia was moved as low and towards the back of the driver head as possible and that the center of gravity remained in a location in line with standard driver heads. The inner lip of the driver head is offset from the outer edge of the acrylic driver face to accommodate the overhang and bend radius of the PVDF sensors from the back of the acrylic driver face.
Figure 7: Driver Head Final Design
Figure 8: Electronic Component Encasement Embedded Mold Final Design
Figure 9: Electronic Component Encapsulation Layers
DAQ
The final data acquisition system consists of three main components, the Teensy 4.0 microcontroller, a digital components printed circuit board (PCB) and an analog components PCB. The Teensy 4.0 microcontroller controls all digital aspects of the circuit, which includes a serial flash memory integrated circuit (IC) for storing collected sensor data, an accelerometer for triggering the sampling of the PVDF force sensors, a radio frequency transciever to wirelessly transmit the collected sensor data, a digital multiplexer to switch between each of the 16 sensors at a rate faster than 200 kSa/s, and programmable gain amplifier for variable signal conditioning levels.
Figure 10: DAQ Components Stack Up
Figure 11: DAQ Block Diagram
The digital components board includes header locations for connecting all the digital components of the system to their respective pins on the Teensy 4.0 microcontroller. All digital components except for the multiplexer operate over the Serial Peripheral Interface (SPI) bus, while the multiplexer is digitally controlled by writing specific pins high (3.3 V) or low (0 V) corresponding to the channel that is to be selected's value in binary. The top of the digital board has outputs for the SPI interface of the programmable gain amplifier and multiplexer to the analog board, as well as an input for the analog signal from the analog board. The digital board connects the accelerometer, radio frequency transciever, and flash memory to the Teensy microcontroller for regular operation.
Figure 12: Digital Components PCB Final Design
The analog board includes headers for the sixteen sensor inputs, which feed into sixteen individual rail-to-rail operational amplifiers, which are configured with a resistor network to take the difference of the two nodes of the sensor. These amplifiers are run by a 3.3 V and 0 V rail, and cut off any signals that are above or below that range, so as to provide protection for the analog to digital converter in the Teensy 4.0 microcontroller that has a reference voltage of 3.3 V. From the differential amplifiers, the signals are fed into a 16 to 1 multiplexer, which switches at rates faster than the aggregate sampling rate of 1 MSa/s. After the multiplexer, another operational amplifier is used to buffer the signal into an optional voltage divider, which is selectable by a solder jumper. The voltage divider gives better flexibility for adjusting the signal range to a minimum of zero volts and a maximum of close to 3.3 V. The gain of this voltage divider can also be adjusted by soldering in different resistors to create the desired gain. In current schematics it is set up as a 1/2 gain voltage divider with two 10k resistors. After the voltage divider is a programmable gain amplifier that can be programmed using SPI through the Teensy. The gain can be adjusted to many different gain levels including 1x, 2x, 4x, 5x, and 8x, but combined with the voltage divider, the overall gain can be more finely tuned. From the programmable gain amplifier are pinouts that connect to the digital board's analog input, and through to the Teensy 4.0's onboard analog to digital converter sampling at 200 kSa/s.
Figure 13: Analog Components PCB Final Design
Power from the lithium ion battery is converted to 5 V in the handle with the Adafruit Powerboost 500c and sent down the shaft to the 5 V inputs of both the analog and digital boards. A 5 V to 3.3 V voltage regulator is used to convert 3.3 V power for running many of the digital components. The Teensy 4.0 is programmable in Arduino IDE, and once programmed, the DAQ is a completely isolated system inside of the club that communicates externally via radio frequency transmission.
Driver Handle
Additional electronic components for data transmission, charging, and user experience were encased inside the 3D printed driver handle. The battery was encased in the driver handle to address safety concerns regarding impact to the battery if it had been placed in the driver head. Positioning the wireless transceiver in the driver handle mitigated the risk of signal loss from the metals typically used to produce driver heads if it had been placed there instead. The on/off switch, battery charger, and LED indicator which indicates on, off, or charging, were also encased in the driver handle.
Figure 14: Driver Handle Assembled View
Figure 15: Driver Handle and Inner Components Exploded View
Driver Shaft
The driver shaft was purchased for product completeness since the initial design comprised of the driver head and driver handle. The purchase shaft was a TaylorMade Fujikura Speeder 65 Regular Flex shaft. Wires connecting components in the driver head and driver handle run inside along the hollow shaft.
Receiver Enclosure
The 3D printed receiver enclosure housed the receiving electronics, and was designed to be small, lightweight, and portable. The wireless transceiver in the driver handle transmits the data to the wireless transceiver located inside the receiver enclosure. The Arduino Mega was directly connected to a laptop, where MATLAB analyzes the data.
Figure 16: Receiver Enclosure and LCD Screen
Figure 17: Receiver Enclosure Inner Components
Receiver Display
The liquid crystal display on the receiver enclosure displays the impact force and location that MATLAB determines. A detailed 3D plot of strike location can be seen on the laptop screen from MATLAB.
Figure 18: Display Setup
Video 1: Receiver Simulation of Randomly Chosen Data
Summary of Performance Results
Full Sensor Layout Validation:
The full sensor layout was validated through the testing with a National Instruments NI PXIe-1082 data acquisition system capable of collecting data from all 16 sensors, given that at the time of testing, the DAQ system for the Smart Golf Clubs was not yet complete. The detection of impact location was achieved through normalization of the data collected and data training analysis.
Theoretical Predictions:
Sensors' output voltages are not proportional to location of golf strike based on a Solidworks simulation where largest deflection of sensor is not correlated to where the golf ball strikes
Normalizing voltage readings from a given impact location will enable a distinction between different impact locations
Sensors' output voltages normalized by its center location output maximum voltage could be compared to all other sensors and can each have a unique ratio of all sensors' outputted voltages for each specific location
A unique identification tag of sensor output voltages for each specific location can be found through data machine learning (data training through hitting multiple trials at each major sensor intersection)
Test Conditions:
Driver face secured to driver head
Force applied using impulse hammer (PCB® Modally Tuned® Impulse Hammer, Model: 086C03)
Variable force applied to driver face at 25 locations specified in Figure 19
10 consecutive impacts at each location
Figure 19: Full Sensor Layout Impact Locations (Major Intersection Spots)
Figure 20: Full Sensor Layout Test Setup
Results:
Some sensors outputted negative voltages, inconsistent data, or no voltage
86.8% test accuracy based out of code correctly calculating simulated data 868 trials out of 1,000 trials although inaccurate calculations determine locations fairly close to the correct location still
Figure 21 shows the average ratio for each sensor at each spot which was calculated with the data training results
Video 1 shows MATLAB correctly calculating the location for a simulated golf strike (4 trials in a row are shown)
Figure 21: Unique Identification Sensor Voltage Ratios for Each Major Intersection Spot
Comparison of Results to Initial Performance Requirements:
Could not perfectly replicate golf strike during experiment because force output voltage would rail
Accuracy of location calibration could be improved by performing more trials per location, adding in more working sensors, and incorporating a method of considering the variable force involved in the location calibration
Figure 21 and the accuracy test result support the notion that data training was effective in characterizing the location of a golf strike
Isolated DAQ Validation:
The isolated DAQ validation was conducted to test if the DAQ could truly sample 16 separate sensor channels at any one time and successfully output that data to individual arrays. To prove that each channel could separately sample individual signals, each channel was set to sample the same signal, and the differences between each channel's output signal to the analog to digital converter were noted. The test was conclusive that each channel could distinctly sample when the same signal was resolved on each channel output, but on a zoomed in scale, small differences could be seen due to delays in sampling time and individual noise on each channel.
Theoretical Predictions:
Each channel outputs an almost identical waveform replicating the 20 Hz sine wave signal
The sine wave signal was attenuated by 1/2 by the voltage divider
Test Conditions:
Sine wave signal output from the National Instruments myDAQ at 20 Hz with a 1 V peak to peak amplitude and a 1 V offset
Signal was outputted to 15 channels on the DAQ, channels 2 through 16
Channel 1 was not fed the 20 Hz signal, leaving it as a control
Analog board was set to sample at the same rate it would measure the PVDF signals, at 5 microseconds per sample, 200 kSa/s aggregate or 12.5 kSa/S per channel
Analog board had its voltage divider enabled, and the programmable gain amplifier was set to 1, so the observed output signal was predicted to be half of the input signal into each of the 16 analog inputs of the analog board
Test setup shown in Figure 22
Input sine wave signal to each of the 15 channels, 2 through 16, is shown in Figure 23 with a 1 V peak to peak amplitude and a 1 V offset
Figure 22: Isolated DAQ Test with Connections to NI myDAQ
Figure 23: Isolated DAQ Test NI myDAQ Output Signal
Results:
Output signal of each channel was sent from the analog PCB and then sampled by the Teensy onboard ADC at 250 kSa/s
Each channel's output is pictured in Figures 24 and 15
Figure 24 shows the full signals, as sampled by the 15 channels and the control at the bottom which was sampling noise
Figure 25 is zoomed in to show the distinction between the different channels as they sample the same signal, due to small switching time delay and individual noise in each channel
Output signal of analog PCB successfully showed a 1/2 gain of the signal
Each sensor input successfully sampled the same signal
The differences in each output signal implied that each channel was truly sampled independently through a different channel on the PCB and through the multiplexer
Figure 24: 20 Hz Sine Wave into 15 Channel Inputs Sampled by Teensy 4.0 ADC
Figure 25: Zoomed in 20 Hz Sine Wave into 15 Channel Inputs Sampled by Teensy 4.0 ADC
Comparison of Results to Initial Performance Requirements:
The DAQ developed for the Smart Clubs project can in fact resolve 16 different signals while maintaining an acceptable sampling rate of above 10 kSa/s
Multiplexer switching time was adequate
The code to separate successive data acquired was successfully implemented to delineate between the stream of data output from the analog PCB into the individual channel arrays
Complete System Validation:
The complete system validation was conducted to verify the sampling capability of the DAQ when connected to the PVDF sensor layout. At the beginning of this test, the primary acrylic driver face failed along a fatigue point very close to the connection points. The fatiguing of the brittle acrylic driver face was not considered as an issue that would significantly affect the plate during testing. Testing was done with the emulation of a 1% strain as the final goal. As such, the sensors were tuned through a resistor network in the DAQ to output in a higher range than the acrylic could ultimately withstand for repeated impacts. The acrylic was impacted many times before final testing of the DAQ was completed, and the useable life of the face expended before the full DAQ could acquire location and force data.
Tests were continued on the fractured acrylic test plate, with full understanding the the percentage strains felt by the PVDF sensors was significantly different than the percentage strains felt under a typical driver face scenario. The PVDF sensors experienced significantly more strain closer to where the fracture occurred no matter where the face was impacted. While any location of force data acquired with the fractured acrylic was not meaningful to data analysis, what could be shown from testing on the fractured acrylic face was the DAQ's ability to read different values from all 16 sensors for different impacts. The tests could also verify that the DAQ could observe variation in the signals, confirming that it could in fact resolve changing voltage outputs from the PVDF sensors.
In continued testing on the fractured acrylic plate, another fracture occurred on the left side of the plate, but substantial data was acquired before the second fracture allowing the PVDF sensor readings on the DAQ to be displayed.
Theoretical Predictions:
The DAQ is capable of sampling the full sensor layout (16 sensors) shown in Figure 26
Validation of DAQ with variations between the values of each sensor for different impacts demonstrates DAQ functionality and can be applied to a newly fabricated PVDF array on an acrylic or titanium driver face
Figure 26: Acrylic Driver Face Fracture
Test Conditions:
Driver face secured to the driver head
Force applied using impulse hammer (PCB® Modally Tuned® Impulse Hammer, Model: 086C03)
Variable force applied to driver face
Results:
Table 1 shows the variations in the sensor output values and control test where no impact was imparted on the face
The values listed are in units of bins of the analog to digital converter
Table 1: Data from Impacts Imparted on Fractured Acrylic Driver Face
Comparison of Results to Initial Performance Requirements:
The varying impact bin values compared to the non-impact maximums clearly indicated that the DAQ could read different voltages presented by the separate sensors
Each sensor responded differently to difference impact forces, impact locations, and amount of sensor flex due to the cracked driver face
Sensor 4 gave very high readings consistently for every location and force value, indicating that it most likely needed a different resistor value for the differential amplifier to match its impedance and bring its overall level lower
Conclusion
The objective of this project was to develop a data acquisition system and sensing strategy capable of capturing the location, force, and attack angle of a golf ball strike on a driver face. The individual components of this project, the DAQ and sensing strategy, succeeded in proving their feasibility of their respective objectives. The Smart Golf Clubs DAQ successfully proved the feasibility of sampling 16 PVDF sensors at a rate of 10 kSa/s. Additionally, the sensing strategy was capable of both localizing and quantifying the force imparted on the driver face via method of calibration and normalization of sensor outputs. Unfortunately, due to the failure of the primary acrylic driver face used for testing, the DAQ and full sensing strategy could not be coupled to form a unified product. Despite this challenge, the success of the individual components of this prototype successfully demonstrated the feasibility of the concept to develop a smart golf club. If this project were to continue, future iterations of the prototype would seek to incorporate a gyroscope and accelerometer combination capable of detecting both the high angular velocities and accelerations typical of a driver swing to enable angle of attack detection. Additionally, future improvements to the DAQ to broaden the detectable range of forces and to decrease its overall size profile would be considered. Apart from the topics involved in producing this prototype, the team learned invaluable lessons on communication and problem solving that was magnified due to the worldly circumstances during which the project was undertaken.
Next Steps
The next steps for this project would need to include a streamlined sensor fabrication process. The by-hand fabrication process implemented for this project would not be ideal for mass production. Additionally, an automated process for encasing the sensor layout in epoxy and for potting the electronic components would aid in mass production. Further testing with this method on materials used in driver production would help verify the accuracy of the downsized force on acrylic analysis and would provide insight on how to proceed with this product. An added feature to better appeal to consumers would be to incorporate a Bluetooth connection option for the receiver enclosure to a laptop or phone. Furthermore, a specialized software incorporated into an application would be beneficial to the consumer as opposed to requiring them to have MATLAB.
The next iteration of the Data Acquisition system could see many changes that can ultimately increase functionality, decrease footprint, and decrease manufacturing time. Including a dedicated analog to digital converter such as the Texas Instruments ADS8686S would increase the ultimate sampling rate and the resolution of the system, while maintaining a much smaller overall footprint than is used in the current analog PCB. Combining the two PCBs is a logical step if enough components are removed and the area of the board is no longer a concern. Moving away from the Teensy 4.0 and solely using the microcontroller integrated circuit chip featured on the Teensy will allow for a large decrease in form factor as well.
Another step to be taken would be to incorporate a gyroscope to pair with an accelerometer. These two components in conjunction would provide an accurate club head location relative to the golfer's swing. This would enable angle detection upon impact and could provide other swing characteristics desirable to a golfer. This process would require developing a gyroscope and accelerometer pairing that could accurately detect movements typical of a golf swing. Today, most inertial measurement units (IMUs) have limits on acceleration or angular velocities that can be detected, with golf swings often surmounting these limits. A future project that was to develop an IMU that could fully capture the characteristics of a golf swing could push this project to the next stage in its development.