|
|
Neural networks on AURIX platform
Smrčka, Michal ; Buchta, Luděk (referee) ; Blaha, Petr (advisor)
Part of this thesis discusses FNNs and how to develop them using MATLAB Deep Learning Toolbox and Keras API in Python. Subsequently, the thesis deals with the conversion of these networks into C/C++ using the Keras2c, AIfES, TFLM and NNoM libraries. This theoretical background was essential for the follow-up section, which focuses on the implementation and testing of the trained FNN on the AURIX TC397 3.3V Application Kit platform. This FNN is used to diagnose the PMS motor in order to detect inter turn faults. The configuration of the GPT12 and GETH peripherals of the AURIX TC397 microcontroller, which were used in the FNN testing application, is described in detail in this thesis. Using the Keras2c library, the possibility of running inference on 2 cores of the AURIX microcontroller was verified and quantization of the trained FNN was performed within the NNoM library. Finally, a comparison of the Keras2c, AIfES, TFLM and NNoM libraries was performed in terms of ease of implementation, classification accuracy and classification speed on the AURIX platform.
|
|
|
Interconnection of AURIX microcontroller with NVIDIA Jetson platform
Smrčka, Michal ; Bartík, Ondřej (referee) ; Blaha, Petr (advisor)
Part of this thesis covers the basics of working with NVIDIA Jetson Nano and AURIX Application Kit platforms, namely TC277 TFT and TC224 TFT. It also theoretically analyzes some peripherals of the platforms. These basics are necessary for the following part, which is focused on the implementation of a serial communication interface for data transfer between platforms. The work describes a program implementation of 2 interfaces: SPI and Ethernet MAC. The communication via SPI (or QSPI in the case of AURIX) is supplemented with the use of GPIO pins, in the case of Ethernet the communication is realized at a link layer using Ethernet frames. The connection via SPI is further tested in a specific application when controlling a BLDC motor using TC224 and AURIX eMotor Drive Kit V2.1, where the measured data is sent to the Jetson Nano for real-time processing.
|
|
|
Neural networks on AURIX platform
Smrčka, Michal ; Buchta, Luděk (referee) ; Blaha, Petr (advisor)
Part of this thesis discusses FNNs and how to develop them using MATLAB Deep Learning Toolbox and Keras API in Python. Subsequently, the thesis deals with the conversion of these networks into C/C++ using the Keras2c, AIfES, TFLM and NNoM libraries. This theoretical background was essential for the follow-up section, which focuses on the implementation and testing of the trained FNN on the AURIX TC397 3.3V Application Kit platform. This FNN is used to diagnose the PMS motor in order to detect inter turn faults. The configuration of the GPT12 and GETH peripherals of the AURIX TC397 microcontroller, which were used in the FNN testing application, is described in detail in this thesis. Using the Keras2c library, the possibility of running inference on 2 cores of the AURIX microcontroller was verified and quantization of the trained FNN was performed within the NNoM library. Finally, a comparison of the Keras2c, AIfES, TFLM and NNoM libraries was performed in terms of ease of implementation, classification accuracy and classification speed on the AURIX platform.
|
|
|
Interconnection of AURIX microcontroller with NVIDIA Jetson platform
Smrčka, Michal ; Bartík, Ondřej (referee) ; Blaha, Petr (advisor)
Part of this thesis covers the basics of working with NVIDIA Jetson Nano and AURIX Application Kit platforms, namely TC277 TFT and TC224 TFT. It also theoretically analyzes some peripherals of the platforms. These basics are necessary for the following part, which is focused on the implementation of a serial communication interface for data transfer between platforms. The work describes a program implementation of 2 interfaces: SPI and Ethernet MAC. The communication via SPI (or QSPI in the case of AURIX) is supplemented with the use of GPIO pins, in the case of Ethernet the communication is realized at a link layer using Ethernet frames. The connection via SPI is further tested in a specific application when controlling a BLDC motor using TC224 and AURIX eMotor Drive Kit V2.1, where the measured data is sent to the Jetson Nano for real-time processing.
|
guest ::
RSS feed.