Squashed 'tinyml-modelzoo/' changes from ab301c33..fd0815a1

fd0815a1 TINYML_ALGO-705
442deb8d Pull request #33: Updated dataset names and removed redundant lines from config
9de8a1f0 Updated dataset names and removed redundant lines from config
61be623e Pull request #32: Feature/SL EDGEAI-45 enable cc2755xxx device support cls 4k npu model and documentation in ccstudio
93057540 SL_EDGEAI-52 - Enable CC13xx devices for Fan Blade
33043973 SL_EDGEAI-45, SL_EDGEAI-59: Add CC2755 and CC35X1 configs for fan blade fault classification
b5c6ac42 Pull request #31: Added 9 classification models and 1 new application (gearbox fault)
6d9594f4 Added error handling and fixed linting issue
7ee77925 Added 9 classification models and 1 new application (gearbox fault)

git-subtree-dir: tinyml-modelzoo
git-subtree-split: fd0815a1bba5a990b619a80afb24dc5960d4185a

Author: Adithya-Thonse

README.md

@@ -109,6 +109,7 @@ tinyml-modelzoo/
 |--------|-----|-------------|-------|
 | CC2755 | CDE | 96 MHz Arm Cortex-M33 wireless MCU | Optimized for PIR/wireless apps |
 | CC1312 | No | Arm Cortex-M4F wireless MCU | Sub-1GHz |
+| CC1314 | No | Arm Cortex-M33 wireless MCU | Sub-1GHz |
 | CC1352 | No | Arm Cortex-M4F wireless MCU | Sub-1GHz |
 | CC1354 | No | Arm Cortex-M33 wireless MCU | Sub-1GHz |
 | CC35X1 | CDE | Arm Cortex-M33 wireless MCU | Wi-Fi + BLE combo |
@@ -259,6 +260,7 @@ These applications are designed for specific use cases with optimized models and
 | **ac_arc_fault** | arc_fault | F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 | Detect AC arc faults in electrical systems |
 | **dc_arc_fault** | arc_fault | F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 | Detect DC arc faults from current waveforms for electrical safety |
 | **ecg_classification** | ecg_classification | MSPM0G3507, MSPM0G5187, MSPM0G3519 | Classify normal vs anomalous heartbeats from ECG signals |
+| **gearbox_fault_detection** | gearbox_fault | MSPM0G3507, MSPM0G3519, MSPM0G5187 | Classify gearbox operating conditions (healthy vs broken tooth) from vibration data |
 | **blower_imbalance** | motor_fault | F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 | Detect blade imbalance in HVAC blowers using 3-phase motor currents |
 | **fan_blade_fault_classification** | motor_fault | F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 | Detect faults in BLDC fans from accelerometer data |
 | **motor_bearing_fault** | motor_fault | F280013, F280015, F28003, F28004, F2837, F28P55, F28P65, MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32, F29H85, AM13E2, AM263 | Classify 5 bearing fault types + normal operation from vibration data |
@@ -268,12 +270,12 @@ These applications are designed for specific use cases with optimized models and
 | **MNIST_image_classification** | image_classification | MSPM0G3507, MSPM0G3519, MSPM0G5187, MSPM33C32 | Handwritten digit recognition (MNIST dataset) |
 
 ### Summary by Task Type:
-- **Generic Timeseries Tasks** (22 examples): Support all target devices and can be adapted to your custom datasets
-  - Classification: 8 examples (1 base + 7 real-world applications)
+- **Generic Timeseries Tasks** (23 examples): Support all target devices and can be adapted to your custom datasets
+  - Classification: 9 examples (1 base + 8 real-world applications)
   - Regression: 5 examples (1 base + 4 real-world applications)
   - Forecasting: 3 examples (1 base + 2 real-world applications)
   - Anomaly Detection: 6 examples (1 base + 5 application variants)
-- **Application-Specific Tasks** (10 examples): arc_fault (2), motor_fault (3), grid_fault_detection (1), mosfet_temp_prediction (1), pir_detection (1), ecg_classification (1), image_classification (1)
+- **Application-Specific Tasks** (11 examples): arc_fault (2), motor_fault (3), grid_fault_detection (1), gearbox_fault_detection (1), mosfet_temp_prediction (1), pir_detection (1), ecg_classification (1), image_classification (1)
 
 ---
 
@@ -288,14 +290,15 @@ These applications are designed for specific use cases with optimized models and
 | 3   | [motor_bearing_fault](examples/motor_bearing_fault/)                                 | Multivariate | Classify 5 bearing fault types + normal operation from vibration data.          |
 | 4   | [blower_imbalance](examples/blower_imbalance/)                                       | Multivariate | Detect blade imbalance in HVAC blowers using 3-phase motor currents.            |
 | 5   | [fan_blade_fault_classification](examples/fan_blade_fault_classification/)           | Multivariate | Detect faults in BLDC fans from accelerometer data.                             |
-| 6   | [electrical_fault](examples/electrical_fault/)                                       | Multivariate | Classify transmission line faults using voltage and current.                    |
-| 7   | [grid_stability](examples/grid_stability/)                                           | Multivariate | Predict power grid stability from node parameters.                              |
-| 8   | [gas_sensor](examples/gas_sensor/)                                                   | Multivariate | Identify gas type and concentration from sensor array data.                     |
-| 9   | [branched_model_parameters](examples/branched_model_parameters/)                     | Multivariate | Human Activity Recognition from accelerometer/gyroscope data.                   |
-| 10  | [ecg_classification](examples/ecg_classification/)                                   | Multivariate | Classify normal vs anomalous heartbeats from ECG signals.                       |
-| 11  | [nilm_appliance_usage_classification](examples/nilm_appliance_usage_classification/) | Multivariate | Non-Intrusive Load Monitoring - identify active appliances.                     |
-| 12  | [PLAID_nilm_classification](examples/PLAID_nilm_classification/)                     | Multivariate | Appliance identification using the PLAID dataset.                               |
-| 13  | [pir_detection](examples/pir_detection/)                                             | Multivariate | Detect presence/motion using PIR sensor data.                                   |
+| 6   | [gearbox_fault_detection](examples/gearbox_fault_detection/)                         | Multivariate | Classify gearbox operating conditions (healthy vs broken tooth) from vibration.  |
+| 7   | [electrical_fault](examples/electrical_fault/)                                       | Multivariate | Classify transmission line faults using voltage and current.                    |
+| 8   | [grid_stability](examples/grid_stability/)                                           | Multivariate | Predict power grid stability from node parameters.                              |
+| 9   | [gas_sensor](examples/gas_sensor/)                                                   | Multivariate | Identify gas type and concentration from sensor array data.                     |
+| 10  | [branched_model_parameters](examples/branched_model_parameters/)                     | Multivariate | Human Activity Recognition from accelerometer/gyroscope data.                   |
+| 11  | [ecg_classification](examples/ecg_classification/)                                   | Multivariate | Classify normal vs anomalous heartbeats from ECG signals.                       |
+| 12  | [nilm_appliance_usage_classification](examples/nilm_appliance_usage_classification/) | Multivariate | Non-Intrusive Load Monitoring - identify active appliances.                     |
+| 13  | [PLAID_nilm_classification](examples/PLAID_nilm_classification/)                     | Multivariate | Appliance identification using the PLAID dataset.                               |
+| 14  | [pir_detection](examples/pir_detection/)                                             | Multivariate | Detect presence/motion using PIR sensor data.                                   |
 
 ### Regression Examples
 
@@ -355,19 +358,30 @@ For detailed guidelines, see [NPU Configuration Guidelines](docs/NPU_CONFIGURATI
 | `CLS_100_NPU` | ~100 | CNN | Yes | Ultra-compact model |
 | `CLS_500_NPU` | ~500 | CNN | Yes | Compact model |
 | `CLS_1k_NPU` | ~1K | CNN | Yes | Lightweight 2-layer CNN |
+| `CLS_1.2k_NPU` | ~1.2K | CNN | Yes | Compact model for ultra-low power devices |
+| `CLS_1.5k_NPU` | ~1.5K | CNN | Yes | 3-layer model with balanced performance |
+| `CLS_1.9k_NPU` | ~1.9K | CNN | Yes | Efficient 3-layer model |
 | `CLS_2k_NPU` | ~2K | CNN | Yes | 2-layer model |
+| `CLS_2.8k_NPU` | ~2.8K | CNN | Yes | Improved accuracy with compact footprint |
+| `CLS_3.1k_NPU` | ~3.1K | CNN | Yes | Higher accuracy model |
 | `CLS_ResAdd_3k` | ~3K | ResNet (Add) | No | Residual connections with addition |
 | `CLS_ResCat_3k` | ~3K | ResNet (Cat) | No | Residual connections with concatenation |
+| `CLS_3.9k_NPU` | ~3.9K | CNN | Yes | Advanced 3-layer model |
 | `CLS_4k_NPU` | ~4K | CNN | Yes | Balanced model |
+| `CLS_4.2k_NPU` | ~4.2K | CNN | Yes | Optimized 4-layer model |
+| `CLS_5k_NPU` | ~5K | CNN | Yes | Mid-range model |
 | `CLS_6k_NPU` | ~6K | CNN (DW-Sep) | Yes | Depthwise separable |
 | `CLS_8k_NPU` | ~8K | CNN (DW-Sep) | Yes | Depthwise separable |
 | `CLS_13k_NPU` | ~13K | CNN | Yes | Higher capacity |
 | `CLS_20k_NPU` | ~20K | CNN | Yes | High capacity |
+| `CLS_40k_NPU` | ~40K | CNN | Yes | Advanced model for complex tasks |
 | `CLS_55k_NPU` | ~55K | CNN | Yes | Maximum accuracy |
 | `ArcFault_model_200_t` | ~200 | Specialized | No | Arc fault detection |
 | `ArcFault_model_300_t` | ~300 | Specialized | No | Arc fault with more capacity |
 | `ArcFault_model_700_t` | ~700 | Specialized | No | Arc fault medium model |
 | `ArcFault_model_1400_t` | ~1.4K | Specialized | No | Arc fault high accuracy |
+| `GearboxFault_model_1.2k_t` | ~1.2K | CNN | Yes | Gearbox fault detection |
+| `GearboxFault_model_1.5k_t` | ~1.5K | CNN | Yes | Gearbox fault with more capacity |
 | `MotorFault_model_1_t` | Varies | Specialized | No | Motor bearing fault detection |
 | `MotorFault_model_2_t` | Varies | Specialized | No | Motor fault variant 2 |
 | `MotorFault_model_3_t` | Varies | Specialized | No | Motor fault variant 3 |

examples/electrical_fault/config_MSPM0.yaml

@@ -0,0 +1,26 @@
+common:
+  task_type: generic_timeseries_classification
+  target_device: MSPM0G5187
+
+dataset:
+  dataset_name: electrical_fault
+  input_data_path: https://software-dl.ti.com/C2000/esd/mcu_ai/01_04_00/datasets/electrical_fault_6class_dataset.zip
+
+data_processing_feature_extraction:
+  data_proc_transforms: ['SimpleWindow']
+  frame_size: 128
+  stride_size: 0.01
+  variables: 3
+  feature_extraction_name: Custom_Default
+  feat_ext_transform: ['FFT_Q15', 'DC_REMOVE', 'Q15_MAG', 'BIN_Q15', 'LOG_DB', 'CONCAT']
+  feature_size_per_frame: 32
+  num_frame_concat: 8
+
+training:
+  model_name: ElectricalFault_model_40k_t
+  training_epochs: 25
+  num_gpus: 0
+
+testing: {}
+
+compilation: {}

examples/electrical_fault/electrical_fault_6class.py

@@ -0,0 +1,104 @@
+import os
+import requests as re
+import zipfile
+import pandas as pd
+import shutil
+
+# Project Definition
+DATASET_NAME = "electrical_fault_6class"
+
+# Dataset definition
+DATASET_URL = "https://software-dl.ti.com/C2000/esd/mcu_ai/01_04_00/datasets/electrical_fault_raw.zip"
+DATASET_FILE_NAME = "classData.csv"
+independent_variables = ["Ia", "Ib", "Ic", "Va", "Vb", "Vc"]
+dependent_variables = ["G", "C", "B", "A"]  # Ground, Node C, Node B, Node A
+
+# Dataset Processing
+NUM_SPLITS = 5
+class_names = [
+    "no_fault",
+    "bc_fault",
+    "abc_fault",
+    "ag_fault",
+    "abg_fault",
+    "abcg_fault",
+]
+
+def download_data():
+    print("Downloading dataset")
+    data = re.get(DATASET_URL)
+    if data.status_code != 200:
+        raise Exception(f"Failed to download dataset. HTTP status code: {data.status_code}")
+    with open(f"{DATASET_NAME}.zip", "wb") as f:
+        f.write(data.content)
+    return None
+
+def extract_data():
+    print("Extracting dataset")
+    with zipfile.ZipFile(f"{DATASET_NAME}.zip", 'r') as zip_ref:
+        zip_ref.extractall(DATASET_NAME)
+    return None
+
+def load_data():
+    data = pd.read_csv(f"{DATASET_NAME}/{DATASET_FILE_NAME}", index_col=False)
+    columns = independent_variables + dependent_variables
+    data = data[columns]
+    # Map the four binary columns to a single target class
+    def map_to_class(row):
+        bits = (row['G'], row['C'], row['B'], row['A'])
+        # Mapping: (G,C,B,A) -> class index
+        if bits == (0,0,0,0):
+            return 0  # NO_FAULT
+        elif bits == (0,1,1,0):
+            return 1  # BC_FAULT
+        elif bits == (0,1,1,1):
+            return 2  # ABC_FAULT
+        elif bits == (1,0,0,1):
+            return 3  # AG_FAULT
+        elif bits == (1,0,1,1):
+            return 4  # ABG_FAULT
+        elif bits == (1,1,1,1):
+            return 5  # ABCG_FAULT
+        else:
+            # Unexpected combination, assign -1 (will become "unknown")
+            return -1
+    data['Target'] = data.apply(map_to_class, axis=1)
+    # Convert to string labels for folder naming
+    data['Target'] = data['Target'].apply(lambda x: class_names[x] if 0 <= x < len(class_names) else "unknown")
+    return data
+
+def store_datafiles(df):
+    if not os.path.exists("classes"):
+        os.mkdir("classes")
+
+    for idx, target in enumerate(class_names):
+        if not os.path.exists(f"classes/class_{idx}_{target}"):
+            os.mkdir(f"classes/class_{idx}_{target}")
+        current_target_df = df[df['Target'] == target]
+        num_splits = NUM_SPLITS
+        chunk_size = len(current_target_df) // num_splits
+        current_target_dfs = [current_target_df[i*chunk_size:(i+1)*chunk_size] for i in range(num_splits)]
+        for df_idx, current_target_df in enumerate(current_target_dfs):
+            current_target_df = current_target_df.drop(['Target'], axis=1)
+            current_target_df.to_csv(f"classes/class_{idx}_{target}/{target}_{df_idx}.csv", index=False)
+        print(f"Created classes/class_{idx}_{target}")
+
+def cleanup():
+    if os.path.exists(DATASET_NAME):
+        shutil.rmtree(DATASET_NAME)
+    zip_path = f"{DATASET_NAME}.zip"
+    if os.path.exists(zip_path):
+        os.remove(zip_path)
+
+def zip_files():
+    print(f"Zipping the classes into {DATASET_NAME}_dataset.zip")    
+    shutil.make_archive(f"{DATASET_NAME}_dataset", 'zip', root_dir='.', base_dir='classes')
+    shutil.rmtree('classes', ignore_errors=True)
+
+if __name__ == '__main__':
+    download_data()
+    extract_data()
+    df = load_data()
+    store_datafiles(df)
+    zip_files()
+    cleanup()

examples/electrical_fault/readme.md

@@ -8,6 +8,10 @@ The Electrical Fault Dataset is a multivariate time series dataset. It is obtain
 
 The fault can be Line-to-line, Line-to-ground, Line-to-line-to-ground and more. Line-to-line (LL) fault is a fault between two phase conductors (e.g., A-B). This typically appears as a short-duration high-energy event affecting particular frequency bins. Line-to-ground (LG) fault: a fault between a phase conductor and ground. Pattern differs from LL faults and can be identified using combinations of voltage and current measurements. LLG / LLLG: multi-conductor faults involving two or three lines and possibly ground. These create distinct signatures across voltage and current channels.
 
+Users can choose between two dataset options, each processed by a different script to produce a distinct output file:
+- **2-class dataset** (`detect_dataset.xlsx`): Processed by `electrical_fault.py` to create `electrical_fault_dataset.zip` for fault detection (binary classification - fault vs no fault)
+- **6-class dataset** (`classData.csv`): Processed by `electrical_fault_6class.py` to create `electrical_fault_6class_dataset.zip` for fault type classification (6 fault types based on G,C,B,A combinations)
+
 <p align="center">  
     <img src="assets/simulink.png" width="280" alt="Simulink Model">
 </p>
@@ -21,26 +25,37 @@ There are two dataset files present in the compressed zip: [electrical_fault_raw
     - There are 4 target variables i.e G (Ground), C (Node C), B (Node B), A (NodeA). The value of each target is either 0 or 1.
     - Examples [G, C, B, A]:
         - [0, 0, 0, 0] means No Fault
+        - [0, 1, 1, 0] means LL Fault btw Node B and Node C
+        - [0, 1, 1, 1] means LLL Fault btw all Nodes
         - [1, 0, 0, 1] means LG Fault btw Ground and Node A
-        - [0, 0, 1, 1] means LL Fault btw Node A and Node B
         - [1, 0, 1, 1] means LLG Fault btw Node A, Node B and Ground
-        - [0, 1, 1, 1] means LLL Fault btw all Nodes
         - [1, 1, 1, 1] means LLLG Fault btw all Nodes and Ground
 
-For this example we will be using `detect_dataset.xlsx` to detect whether there is electrical fault or not.
+Depending on the dataset chosen, users can either detect whether there is an electrical fault (binary classification) or classify the type of fault (6-class classification).
 
 ## Downloading dataset
 
-Prepare the zipped dataset by running the electrical_fault python file. The script will create zipped dataset as `electrical_fault_dataset.zip`. 
+Users can prepare the zipped dataset using either of two python scripts, depending on which dataset they want to use:
+
+**For 2-class dataset (fault detection):**
 ```bash
 cd examples/electrical_fault
 python electrical_fault.py
 ```
-The path of this zipped dataset file is already mentioned in [configuration](config.yaml) yaml, make sure it is same.
+This creates `electrical_fault_dataset.zip`
+
+**For 6-class dataset (fault type classification):**
+```bash
+cd examples/electrical_fault
+python electrical_fault_6class.py
+```
+This creates `electrical_fault_6class_dataset.zip`
+
+The path to the appropriate zipped dataset file should be mentioned in [configuration](config.yaml) yaml under `dataset.input_data_path`, make sure it matches the script you ran.
 
 ```yaml
 dataset:
-    input_data_path: 'examples/electrical_fault/electrical_fault_dataset.zip'
+    input_data_path: 'examples/electrical_fault/electrical_fault_dataset.zip'   # OR 'examples/electrical_fault/electrical_fault_6class_dataset.zip'
 ```
 
 ## Usage in Tiny ML ModelZoo

examples/fan_blade_fault_classification/config_CC1312.yaml

@@ -0,0 +1,34 @@
+common:
+  task_type: motor_fault
+  target_device: CC1312
+dataset:
+  dataset_name: fan_blade_fault
+  input_data_path: https://software-dl.ti.com/C2000/esd/mcu_ai/01_04_00/datasets/fan_blade_fault_dsi.zip
+data_processing_feature_extraction:
+  data_proc_transforms: []
+  feature_extraction_name: Input256_FFTBIN_16Feature_8Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Input256_FFTBIN_16Feature_8Frame_3InputChannel_removeDC_1D
+#    feature_extraction_name: Input256_FFT_128Feature_1Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Input128_RAW_128Feature_1Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Custom_MotorFault
+#    frame_size: 256
+#    feature_size_per_frame: 128
+#    num_frame_concat: 8
+#    normalize_bin: 1
+#    stacking: 2D1
+#    feat_ext_transform: [ 'FFT_FE', 'FFT_POS_HALF', 'DC_REMOVE', 'ABS', 'LOG_DB', 'CONCAT' ]
+#    offset: 0
+#    scale: 1
+#    frame_skip: 1
+#    log_mul: 20
+#    log_base: 10
+#    log_threshold: 1e-100
+  variables: 3
+training:
+  model_name: CLS_4k_NPU
+  model_config: ''
+  batch_size: 256
+  training_epochs: 20
+  num_gpus: 0
+testing: {}
+compilation: {}

examples/fan_blade_fault_classification/config_CC1314.yaml

@@ -0,0 +1,34 @@
+common:
+  task_type: motor_fault
+  target_device: CC1314
+dataset:
+  dataset_name: fan_blade_fault
+  input_data_path: https://software-dl.ti.com/C2000/esd/mcu_ai/01_04_00/datasets/fan_blade_fault_dsi.zip
+data_processing_feature_extraction:
+  data_proc_transforms: []
+  feature_extraction_name: Input256_FFTBIN_16Feature_8Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Input256_FFTBIN_16Feature_8Frame_3InputChannel_removeDC_1D
+#    feature_extraction_name: Input256_FFT_128Feature_1Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Input128_RAW_128Feature_1Frame_3InputChannel_removeDC_2D1
+#    feature_extraction_name: Custom_MotorFault
+#    frame_size: 256
+#    feature_size_per_frame: 128
+#    num_frame_concat: 8
+#    normalize_bin: 1
+#    stacking: 2D1
+#    feat_ext_transform: [ 'FFT_FE', 'FFT_POS_HALF', 'DC_REMOVE', 'ABS', 'LOG_DB', 'CONCAT' ]
+#    offset: 0
+#    scale: 1
+#    frame_skip: 1
+#    log_mul: 20
+#    log_base: 10
+#    log_threshold: 1e-100
+  variables: 3
+training:
+  model_name: CLS_4k_NPU
+  model_config: ''
+  batch_size: 256
+  training_epochs: 20
+  num_gpus: 0
+testing: {}
+compilation: {}