Update app.py
Browse files
app.py
CHANGED
@@ -18,140 +18,195 @@ DEFAULT_THRESHOLD = 0.7
|
|
18 |
feature_extractor = AutoFeatureExtractor.from_pretrained(MODEL_NAME)
|
19 |
model = AutoModelForAudioClassification.from_pretrained(MODEL_NAME)
|
20 |
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
-
|
30 |
-
|
31 |
|
32 |
-
|
33 |
-
|
34 |
-
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
39 |
|
40 |
def analyze_audio(audio_input, threshold=DEFAULT_THRESHOLD):
|
41 |
"""Process audio and detect anomalies"""
|
42 |
try:
|
43 |
-
# Handle
|
44 |
-
if isinstance(audio_input, str):
|
45 |
-
audio, sr =
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
|
53 |
-
#
|
|
|
|
|
54 |
if sr != SAMPLING_RATE:
|
55 |
audio = librosa.resample(audio, orig_sr=sr, target_sr=SAMPLING_RATE)
|
56 |
|
57 |
-
#
|
58 |
-
inputs = feature_extractor(
|
59 |
-
audio,
|
60 |
-
sampling_rate=SAMPLING_RATE,
|
61 |
-
return_tensors="pt",
|
62 |
-
padding=True,
|
63 |
-
return_attention_mask=True
|
64 |
-
)
|
65 |
-
|
66 |
-
# Run inference
|
67 |
with torch.no_grad():
|
68 |
outputs = model(**inputs)
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
# Get results
|
73 |
predicted_class = "Normal" if probs[0][0] > threshold else "Anomaly"
|
74 |
confidence = probs[0][0].item() if predicted_class == "Normal" else 1 - probs[0][0].item()
|
75 |
|
76 |
-
#
|
77 |
-
spectrogram = librosa.feature.melspectrogram(
|
78 |
-
y=audio,
|
79 |
-
sr=SAMPLING_RATE,
|
80 |
-
n_mels=64,
|
81 |
-
fmax=8000
|
82 |
-
)
|
83 |
db_spec = librosa.power_to_db(spectrogram, ref=np.max)
|
84 |
|
85 |
fig, ax = plt.subplots(figsize=(10, 4))
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
ax.set(title='Mel Spectrogram')
|
96 |
-
plt.tight_layout()
|
97 |
|
98 |
-
# Save to temp file
|
99 |
spec_path = os.path.join(tempfile.gettempdir(), 'spec.png')
|
100 |
plt.savefig(spec_path, bbox_inches='tight')
|
101 |
plt.close()
|
102 |
|
|
|
|
|
|
|
103 |
return (
|
104 |
predicted_class,
|
105 |
f"{confidence:.1%}",
|
106 |
spec_path,
|
107 |
-
|
108 |
)
|
109 |
|
110 |
except Exception as e:
|
111 |
return f"Error: {str(e)}", "", None, ""
|
112 |
|
113 |
-
# Gradio
|
114 |
-
with gr.Blocks(title="Industrial
|
115 |
gr.Markdown("""
|
116 |
-
# π Industrial Equipment
|
117 |
-
|
118 |
""")
|
119 |
|
120 |
with gr.Row():
|
121 |
with gr.Column():
|
122 |
audio_input = gr.Audio(
|
123 |
-
label="Upload Equipment
|
124 |
-
type="filepath"
|
|
|
125 |
)
|
126 |
threshold = gr.Slider(
|
127 |
-
minimum=0.5,
|
128 |
-
|
129 |
-
step=0.05,
|
130 |
-
value=DEFAULT_THRESHOLD,
|
131 |
-
label="Anomaly Detection Threshold"
|
132 |
)
|
133 |
-
analyze_btn = gr.Button("π Analyze
|
134 |
|
135 |
with gr.Column():
|
136 |
-
result_label = gr.Label(label="
|
137 |
confidence = gr.Textbox(label="Confidence Score")
|
138 |
-
spectrogram = gr.Image(label="
|
139 |
-
|
140 |
-
label="
|
141 |
-
|
|
|
142 |
)
|
143 |
|
144 |
analyze_btn.click(
|
145 |
fn=analyze_audio,
|
146 |
inputs=[audio_input, threshold],
|
147 |
-
outputs=[result_label, confidence, spectrogram,
|
148 |
)
|
149 |
|
150 |
gr.Markdown("""
|
151 |
-
|
152 |
-
-
|
153 |
-
-
|
154 |
-
-
|
|
|
|
|
|
|
155 |
""")
|
156 |
|
157 |
if __name__ == "__main__":
|
|
|
18 |
feature_extractor = AutoFeatureExtractor.from_pretrained(MODEL_NAME)
|
19 |
model = AutoModelForAudioClassification.from_pretrained(MODEL_NAME)
|
20 |
|
21 |
+
# Equipment knowledge base
|
22 |
+
EQUIPMENT_RECOMMENDATIONS = {
|
23 |
+
"bearing": {
|
24 |
+
"high_frequency": "Recommend bearing replacement. High-frequency noise indicates wear or lubrication issues.",
|
25 |
+
"low_frequency": "Check for improper installation or contamination in bearings.",
|
26 |
+
"irregular": "Possible bearing cage damage. Schedule vibration analysis."
|
27 |
+
},
|
28 |
+
"pump": {
|
29 |
+
"cavitation": "Pump cavitation detected. Check suction conditions and NPSH.",
|
30 |
+
"impeller": "Impeller damage likely. Inspect and balance if needed.",
|
31 |
+
"misalignment": "Misalignment detected. Perform laser shaft alignment."
|
32 |
+
},
|
33 |
+
"motor": {
|
34 |
+
"electrical": "Electrical fault suspected. Check windings and connections.",
|
35 |
+
"mechanical": "Mechanical imbalance detected. Perform dynamic balancing.",
|
36 |
+
"bearing": "Motor bearing wear detected. Schedule replacement."
|
37 |
+
},
|
38 |
+
"compressor": {
|
39 |
+
"valve": "Compressor valve leakage suspected. Perform valve test.",
|
40 |
+
"pulsation": "Pulsation issues detected. Check dampeners and piping.",
|
41 |
+
"surge": "Compressor surge condition. Review control settings."
|
42 |
+
}
|
43 |
+
}
|
44 |
+
|
45 |
+
def analyze_frequency_patterns(audio, sr):
|
46 |
+
"""Analyze frequency patterns to identify potential issues"""
|
47 |
+
patterns = []
|
48 |
+
|
49 |
+
# Spectral analysis
|
50 |
+
spectral_centroid = librosa.feature.spectral_centroid(y=audio, sr=sr)[0]
|
51 |
+
spectral_rolloff = librosa.feature.spectral_rolloff(y=audio, sr=sr)[0]
|
52 |
+
|
53 |
+
mean_centroid = np.mean(spectral_centroid)
|
54 |
+
mean_rolloff = np.mean(spectral_rolloff)
|
55 |
+
|
56 |
+
if mean_centroid > 3000: # High frequency components
|
57 |
+
patterns.append("high_frequency")
|
58 |
+
elif mean_centroid < 1000: # Low frequency components
|
59 |
+
patterns.append("low_frequency")
|
60 |
|
61 |
+
if mean_rolloff > 8000: # Rich in harmonics
|
62 |
+
patterns.append("harmonic_rich")
|
63 |
|
64 |
+
return patterns
|
65 |
+
|
66 |
+
def generate_recommendation(prediction, confidence, audio, sr):
|
67 |
+
"""Generate maintenance recommendations based on analysis"""
|
68 |
+
if prediction == "Normal":
|
69 |
+
return "No immediate action required. Equipment operating within normal parameters."
|
70 |
+
|
71 |
+
patterns = analyze_frequency_patterns(audio, sr)
|
72 |
+
|
73 |
+
# Simple equipment type classifier based on frequency profile
|
74 |
+
spectral_flatness = librosa.feature.spectral_flatness(y=audio)[0]
|
75 |
+
mean_flatness = np.mean(spectral_flatness)
|
76 |
+
|
77 |
+
if mean_flatness < 0.2:
|
78 |
+
equipment_type = "bearing"
|
79 |
+
elif 0.2 <= mean_flatness < 0.6:
|
80 |
+
equipment_type = "pump"
|
81 |
+
else:
|
82 |
+
equipment_type = "motor" if np.mean(audio) < 0.1 else "compressor"
|
83 |
+
|
84 |
+
# Generate specific recommendations
|
85 |
+
recommendations = ["π§ Maintenance Recommendations:"]
|
86 |
+
recommendations.append(f"Detected issues in {equipment_type} with {confidence:.1%} confidence")
|
87 |
+
|
88 |
+
for pattern in patterns:
|
89 |
+
if pattern in EQUIPMENT_RECOMMENDATIONS.get(equipment_type, {}):
|
90 |
+
recommendations.append(f"β {EQUIPMENT_RECOMMENDATIONS[equipment_type][pattern]}")
|
91 |
+
|
92 |
+
# General recommendations
|
93 |
+
if prediction == "Anomaly":
|
94 |
+
recommendations.append("\nπ οΈ Suggested Actions:")
|
95 |
+
recommendations.append("1. Isolate equipment if possible")
|
96 |
+
recommendations.append("2. Perform visual inspection")
|
97 |
+
recommendations.append("3. Schedule detailed diagnostics")
|
98 |
+
recommendations.append(f"4. Review last maintenance records ({equipment_type})")
|
99 |
+
|
100 |
+
if confidence > 0.8:
|
101 |
+
recommendations.append("\nπ¨ Urgent: High confidence abnormality detected. Recommend immediate inspection!")
|
102 |
+
|
103 |
+
return "\n".join(recommendations)
|
104 |
|
105 |
def analyze_audio(audio_input, threshold=DEFAULT_THRESHOLD):
|
106 |
"""Process audio and detect anomalies"""
|
107 |
try:
|
108 |
+
# Handle file upload
|
109 |
+
if isinstance(audio_input, str):
|
110 |
+
audio, sr = sf.read(audio_input)
|
111 |
+
else: # Gradio file object
|
112 |
+
with tempfile.NamedTemporaryFile(suffix='.wav', delete=False) as tmp:
|
113 |
+
tmp.write(audio_input.read())
|
114 |
+
tmp_path = tmp.name
|
115 |
+
audio, sr = sf.read(tmp_path)
|
116 |
+
os.unlink(tmp_path)
|
117 |
|
118 |
+
# Convert to mono and resample if needed
|
119 |
+
if len(audio.shape) > 1:
|
120 |
+
audio = np.mean(audio, axis=1)
|
121 |
if sr != SAMPLING_RATE:
|
122 |
audio = librosa.resample(audio, orig_sr=sr, target_sr=SAMPLING_RATE)
|
123 |
|
124 |
+
# Feature extraction and prediction
|
125 |
+
inputs = feature_extractor(audio, sampling_rate=SAMPLING_RATE, return_tensors="pt")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
with torch.no_grad():
|
127 |
outputs = model(**inputs)
|
128 |
+
probs = torch.softmax(outputs.logits, dim=-1)
|
129 |
+
|
|
|
130 |
# Get results
|
131 |
predicted_class = "Normal" if probs[0][0] > threshold else "Anomaly"
|
132 |
confidence = probs[0][0].item() if predicted_class == "Normal" else 1 - probs[0][0].item()
|
133 |
|
134 |
+
# Generate spectrogram
|
135 |
+
spectrogram = librosa.feature.melspectrogram(y=audio, sr=SAMPLING_RATE, n_mels=64, fmax=8000)
|
|
|
|
|
|
|
|
|
|
|
136 |
db_spec = librosa.power_to_db(spectrogram, ref=np.max)
|
137 |
|
138 |
fig, ax = plt.subplots(figsize=(10, 4))
|
139 |
+
librosa.display.specshow(db_spec, x_axis='time', y_axis='mel', sr=SAMPLING_RATE, fmax=8000, ax=ax)
|
140 |
+
plt.colorbar(format='%+2.0f dB')
|
141 |
+
plt.title('Mel Spectrogram with Anomaly Detection')
|
142 |
+
|
143 |
+
# Mark anomalies on plot
|
144 |
+
if predicted_class == "Anomaly":
|
145 |
+
plt.text(0.5, 0.9, 'ANOMALY DETECTED', color='red',
|
146 |
+
ha='center', va='center', transform=ax.transAxes,
|
147 |
+
fontsize=14, bbox=dict(facecolor='white', alpha=0.8))
|
|
|
|
|
148 |
|
|
|
149 |
spec_path = os.path.join(tempfile.gettempdir(), 'spec.png')
|
150 |
plt.savefig(spec_path, bbox_inches='tight')
|
151 |
plt.close()
|
152 |
|
153 |
+
# Generate detailed recommendations
|
154 |
+
recommendations = generate_recommendation(predicted_class, confidence, audio, SAMPLING_RATE)
|
155 |
+
|
156 |
return (
|
157 |
predicted_class,
|
158 |
f"{confidence:.1%}",
|
159 |
spec_path,
|
160 |
+
recommendations
|
161 |
)
|
162 |
|
163 |
except Exception as e:
|
164 |
return f"Error: {str(e)}", "", None, ""
|
165 |
|
166 |
+
# Gradio Interface
|
167 |
+
with gr.Blocks(title="Industrial Diagnostic Assistant π¨βπ§", theme=gr.themes.Soft()) as demo:
|
168 |
gr.Markdown("""
|
169 |
+
# π Industrial Equipment Diagnostic Assistant
|
170 |
+
## Acoustic Anomaly Detection & Maintenance Recommendation System
|
171 |
""")
|
172 |
|
173 |
with gr.Row():
|
174 |
with gr.Column():
|
175 |
audio_input = gr.Audio(
|
176 |
+
label="Upload Equipment Recording (.wav)",
|
177 |
+
type="filepath",
|
178 |
+
source="upload"
|
179 |
)
|
180 |
threshold = gr.Slider(
|
181 |
+
minimum=0.5, maximum=0.95, step=0.05, value=DEFAULT_THRESHOLD,
|
182 |
+
label="Detection Sensitivity", interactive=True
|
|
|
|
|
|
|
183 |
)
|
184 |
+
analyze_btn = gr.Button("π Analyze & Diagnose", variant="primary")
|
185 |
|
186 |
with gr.Column():
|
187 |
+
result_label = gr.Label(label="Diagnosis Result")
|
188 |
confidence = gr.Textbox(label="Confidence Score")
|
189 |
+
spectrogram = gr.Image(label="Acoustic Analysis")
|
190 |
+
recommendations = gr.Textbox(
|
191 |
+
label="Maintenance Recommendations",
|
192 |
+
lines=10,
|
193 |
+
interactive=False
|
194 |
)
|
195 |
|
196 |
analyze_btn.click(
|
197 |
fn=analyze_audio,
|
198 |
inputs=[audio_input, threshold],
|
199 |
+
outputs=[result_label, confidence, spectrogram, recommendations]
|
200 |
)
|
201 |
|
202 |
gr.Markdown("""
|
203 |
+
### System Capabilities:
|
204 |
+
- Automatic anomaly detection in industrial equipment sounds
|
205 |
+
- Frequency pattern analysis to identify failure modes
|
206 |
+
- Equipment-specific maintenance recommendations
|
207 |
+
- Confidence-based urgency classification
|
208 |
+
|
209 |
+
**Tip:** For best results, use 5-10 second recordings of steady operation
|
210 |
""")
|
211 |
|
212 |
if __name__ == "__main__":
|