This document describes the process of setting up and running the Arm® Ethos™-U NPU Anomaly Detection example.
Use-case code could be found in the following directory: source/use_case/ad.
The Anomaly Detection model that is used with the Code Samples and expects audio data to be preprocessed in a specific way before performing an inference.
Therefore, this section provides an overview of the feature extraction process used.
First, the audio data is normalized to the range (-1, 1).
Next, a window of 1024 audio samples is taken from the start of the audio clip. From these 1024 samples, we calculate 64 Log Mel Energies that form part of a Log Mel Spectrogram.
The window is shifted by 512 audio samples and another 64 Log Mel Energies are calculated. This is repeated until we have 64 sets of Log Mel Energies.
This 64x64 matrix of values is then resized by a factor of two, resulting in a 32x32 matrix of values.
The average of the training dataset is then subtracted from this 32x32 matrix and an inference can now be performed.
We start this process again, but shift the start by 20*512=10240 audio samples. This keeps repeating until enough inferences have been performed to cover the whole audio clip.
Softmax is then applied to the result of each inference. Based on the machine ID of the wav clip being processed, we look at a specific index in each output vector. An average of the negative value at this index across all the inferences performed for the audio clip is taken.
If this average value is greater than a chosen threshold score, then the machine in the clip is not behaving anomalously. If the score is lower than the threshold, then the machine in the clip is behaving anomalously.
See Prerequisites
In addition to the already specified build option in the main documentation, the Anomaly Detection use-case adds:
ad_MODEL_TFLITE_PATH - Path to the NN model file in the TFLite format. The model is then processed and included in the application axf file. The default value points to one of the delivered set of models.
Note that the parameters ad_LABELS_TXT_FILE, TARGET_PLATFORM, and ETHOS_U_NPU_ENABLED must be aligned with the chosen model. In other words:
ETHOS_U_NPU_ENABLED is set to On or 1, then the NN model is assumed to be optimized. The model naturally falls back to the Arm® Cortex®-M CPU if an unoptimized model is supplied.ETHOS_U_NPU_ENABLED is set to Off or 0, the NN model is assumed to be unoptimized. Supplying an optimized model in this case results in a runtime error.ad_FILE_PATH: Path to the directory containing audio files, or a path to single WAV file, to be used in the application. The default value points to the resources/ad/samples folder containing the delivered set of audio clips.
ad_AUDIO_RATE: The input data sampling rate. Each audio file from ad_FILE_PATH is preprocessed during the build to match the NN model input requirements. The default value is 16000.
ad_AUDIO_MONO: If set to ON, then the audio data is converted to mono. The default value is ON.
ad_AUDIO_OFFSET: begin loading the audio data, while starting from this offset amount, defined in seconds. The default value is set to 0.
ad_AUDIO_DURATION: Length of the audio data to be used in the application in seconds. Default is 0, meaning that the whole audio file is used.
ad_AUDIO_MIN_SAMPLES: Minimum number of samples required by the network model. If the audio clip is shorter than this number, then it is padded with zeros. The default value is 16000.
ad_MODEL_SCORE_THRESHOLD: Threshold value to be applied to average Softmax score over the clip, if larger than this value, then there is an anomaly.
ad_ACTIVATION_BUF_SZ: The intermediate, or activation, buffer size reserved for the NN model. By default, it is set to 2MiB and is enough for most models.
In order to ONLY build the Anomaly Detection example application, add -DUSE_CASE_BUILD=ad to the cmake command line that is specified in: Building.
Note: This section describes the process for configuring the build for the
MPS3: SSE-300for a different target platform. Additional information can be found at: Building.
Create a build directory folder and then navigate inside using:
mkdir build_ad && cd build_ad
On Linux, when providing only the mandatory arguments for CMake configuration, execute the following command to only build the Anomaly Detection application to run on the Ethos-U55 Fast Model:
cmake ../ -DUSE_CASE_BUILD=ad
To configure a build that can be debugged using Arm DS, specify the build type as Debug and use the Arm Compiler toolchain file, like so:
cmake .. \ -DCMAKE_TOOLCHAIN_FILE=scripts/cmake/toolchains/bare-metal-armclang.cmake \ -DCMAKE_BUILD_TYPE=Debug \ -DUSE_CASE_BUILD=ad
For additional information, please refer to:
Note: If re-building with changed parameters values, we recommend that you clean the build directory and then re-run the CMake command.
If the CMake command succeeded, build the application as follows:
make -j4
To see compilation and link details, add VERBOSE=1.
Results of the build are placed under build/bin folder. For example:
bin ├── ethos-u-.axf ├── ethos-u-ad.htm ├── ethos-u-.map └── sectors ├── images.txt └── ad ├── ddr.bin └── itcm.bin
The bin folder contains the following files and folders:
ethos-u-ad.axf: The built application binary for the Anomaly Detection use-case.
ethos-u-ad.map: Information from building the application. For example, the libraries used, what was optimized, and the location of objects.
ethos-u-ad.htm: Human readable file containing the call graph of application functions.
sectors/ad: Folder containing the built application. is split into files for loading into different FPGA memory regions.
sectors/images.txt: Tells the FPGA which memory regions to use for loading the binaries in the sectors/.. folder.
The application anomaly detection is set up to perform inferences on data found in the folder, or an individual file, that is pointed to by the parameter ad_FILE_PATH.
To run the application with your own audio clips, first create a folder to hold them and then copy the custom clips into the following folder:
mkdir /tmp/custom_files cp custom_id_00.wav /tmp/custom_files/
Note: The data used for this example comes from https://zenodo.org/record/3384388#.X6GILFNKiqA and the model included in this example is trained on the "Slider" part of the dataset.
The machine ID for the clip, so:00,02,04,06, comes from must be in the file name for the application to work.
The file name must have a pattern that matches. For example:<any>_<text>_00_<here>.wavif the audio was from machine ID00, or<any>_<text>_02_<here>.wavif it was from machine ID02, and so on.Note: Clean the build directory before re-running the CMake command.
Next, set ad_FILE_PATH to the location of the following folder when building:
cmake .. \ -Dad_FILE_PATH=/tmp/custom_files/ \ -DUSE_CASE_BUILD=ad
The audio flies found in the ad_FILE_PATH folder are picked up and automatically converted to C++ files during the CMake configuration stage. They are then compiled into the application during the build phase for performing inference with.
The log from the configuration stage tells you what image directory path has been used:
-- User option ad_FILE_PATH is set to /tmp/custom_files
After compiling, your custom inputs have now replaced the default ones in the application.
The application performs inference using the model pointed to by the CMake parameter ad_MODEL_TFLITE_PATH.
Note: If you want to run the model using an Ethos-U, ensure that your custom model has been successfully run through the Vela compiler before continuing. Please refer to this section for more help: Optimize model with Vela compiler.
For example:
cmake .. \ -Dad_MODEL_TFLITE_PATH=<path/to/custom_ad_model_after_vela.tflite> \ -DUSE_CASE_BUILD=ad
Note: Clean the build directory before re-running the CMake command.
The .tflite model file pointed to by ad_MODEL_TFLITE_PATH is converted to C++ files during the CMake configuration stage and is then compiled into the application for performing inference with.
The log from the configuration stage tells you what model path has been used. For example:
-- User option TARGET_PLATFORM is set to fastmodel -- User option ad_MODEL_TFLITE_PATH is set to <path/to/custom_ad_model_after_vela.tflite> ... -- Using <path/to/custom_ad_model_after_vela.tflite> ++ Converting custom_ad_model_after_vela.tflite to custom_ad_model_after_vela.tflite.cc ...
After compiling, your custom model has now replaced the default one in the application.
Note: To successfully run the model, the NPU must be enabled and the platform
TARGET_PLATFORMis set tomps3andTARGET_SUBSYSTEMisSSE-300.
The FVP is available publicly from Arm Ecosystem FVP downloads.
For the Ethos-U evaluation, please download the MPS3 based version of the Arm® Corstone™-300 model that contains Cortex-M55 and offers a choice of the Ethos-U55 and Ethos-U65 processors.
To install the FVP:
Unpack the archive.
Run the install script in the extracted package:
./FVP_Corstone_SSE-300.sh
Note: The anomaly detection example does not come pre-built. Therefore, you must first follow the instructions in section three for building the application from source.
After building, and assuming the install location of the FVP was set to the ~/FVP_install_location, the simulation can be started by running:
~/FVP_install_location/models/Linux64_GCC-6.4/FVP_Corstone_SSE-300_Ethos-U55 ./bin/ethos-u-ad.axf
A log output now appears on the terminal:
telnetterminal0: Listening for serial connection on port 5000 telnetterminal1: Listening for serial connection on port 5001 telnetterminal2: Listening for serial connection on port 5002 telnetterminal5: Listening for serial connection on port 5003
This also launches a telnet window with the standard output of the sample application. It also includes error log entries containing information about the pre-built application version, TensorFlow Lite Micro library version used, and data types. The log also includes the input and output tensor sizes of the model compiled into the executable binary.
After the application has started, if ad_FILE_PATH points to a single file, or even a folder that contains a single input file, then the inference starts immediately. If there are multiple inputs, it outputs a menu and then waits for input from the user. For example:
User input required Enter option number from: 1. Classify next audio signal 2. Classify audio signal at chosen index 3. Run classification on all audio signals 4. Show NN model info 5. List audio signals Choice:
What the preceding choices do:
Classify next audio clip: Runs a single inference on the next in line.
Classify audio clip at chosen index: Runs inference on the chosen audio clip.
Note: Please make sure to select audio clip index within the range of supplied audio clips during application build. By default, the pre-built application has one file with index
0.
Run ... on all: Triggers sequential inference executions on all built-in applications.
Show NN model info: Prints information about the model data type, input, and output, tensor sizes:
INFO - Model info: INFO - Model INPUT tensors: INFO - tensor type is INT8 INFO - tensor occupies 1024 bytes with dimensions INFO - 0: 1 INFO - 1: 32 INFO - 2: 32 INFO - 3: 1 INFO - Quant dimension: 0 INFO - Scale[0] = 0.192437 INFO - ZeroPoint[0] = 11 INFO - Model OUTPUT tensors: INFO - tensor type is INT8 INFO - tensor occupies 8 bytes with dimensions INFO - 0: 1 INFO - 1: 8 INFO - Quant dimension: 0 INFO - Scale[0] = 0.048891 INFO - ZeroPoint[0] = -30 INFO - Activation buffer (a.k.a tensor arena) size used: 321252 INFO - Number of operators: 1 INFO - Operator 0: ethos-u
List: Prints a list of pair ... indexes. The original filenames are embedded in the application, like so:
INFO - List of Files: INFO - 0 =>; random_id_00_000000.wav
Please select the first menu option to execute the Anomaly Detection.
The following example illustrates the output of an application:
INFO - Running inference on audio clip 4 => random_id_00_000000.wav INFO - Inference 1/1 INFO - Average anomaly score is: -0.883147 INFO - Anomaly threshold is: -0.800000 INFO - Everything fine, no anomaly detected! INFO - Profile for Inference: INFO - NPU AXI0_RD_DATA_BEAT_RECEIVED beats: 485194 INFO - NPU AXI0_WR_DATA_BEAT_WRITTEN beats: 138111 INFO - NPU AXI1_RD_DATA_BEAT_RECEIVED beats: 56245 INFO - NPU ACTIVE cycles: 967731 INFO - NPU IDLE cycles: 441 INFO - NPU TOTAL cycles: 968172
As multiple inferences must be run for one clip, it takes around a minute for all inferences to complete.
For the random_id_00_000000.wav clip, after averaging results across all inferences needed, the score is less than the chosen anomaly threshold. Therefore, an anomaly was not detected with the machine in this clip.
The profiling section of the log shows that for each inference. For the last inference, the profiling reports:
Ethos-U PMU report:
968,172 total cycle: The number of NPU cycles
967,731 active cycles: number of NPU cycles that were used for computation
441 idle cycles: number of cycles for which the NPU was idle
48,5194 AXI0 read beats: The number of AXI beats with read transactions from AXI0 bus. AXI0 is the bus where Ethos-U NPU reads and writes to the computation buffers, activation buf, or tensor arenas.
138,111 AXI0 write beats: The number of AXI beats with write transactions to AXI0 bus.
56,245 AXI1 read beats: The number of AXI beats with read transactions from AXI1 bus. AXI1 is the bus where Ethos-U NPU reads the model. So, read-only.
For FPGA platforms, a CPU cycle count can also be enabled. However, do not use cycle counters for FVP, as the CPU model is not cycle-approximate or cycle-accurate.