app.py

import streamlit as st
import numpy as np
import matplotlib.pyplot as plt

# --------------------------------------
# Utility functions
# --------------------------------------

def generate_random_pile(num_points=50, max_height=10.0, noise=0.5):
    """
    Generates a random 'pile shape' with a certain maximum height.
    shape it as a bell-curve-like mound plus random noise.
    
    Args:
        num_points (int): Number of discrete x-points to model the silo cross-section.
        max_height (float): Approximate max possible height.
        noise (float): Random fluctuations around the main shape.
        
    Returns:
        x_array (np.array): X-coordinates from 0..1 (normalized).
        height_array (np.array): Random mound shape values at each x.
    """
    x_array = np.linspace(0, 1, num_points)
    center = 0.5
    sigma = 0.15  # controls how wide the mound is
    base_curve = max_height * np.exp(-0.5 * ((x_array - center) / sigma)**2)
    
    random_variation = noise * np.random.randn(num_points)
    height_array = np.clip(base_curve + random_variation, 0, None)
    return x_array, height_array

def simulate_sensor_readings(x_array, height_array, sensor_type="LiDAR", num_sensors=5):
    """
    Simulates sensor readings from the height array.
    randomly pick a few positions along x_array,
    then record the height + some noise depending on sensor_type.
    
    Args:
        x_array (np.array): X-coordinates for the silo cross-section
        height_array (np.array): True pile height at each x
        sensor_type (str): "LiDAR" or "Ultrasonic" - determines noise level
        num_sensors (int): number of sensors/points to read
        
    Returns:
        sensor_positions (np.array): Chosen positions
        sensor_heights (np.array): Measured heights (with noise)
    """
    sensor_positions = np.random.choice(x_array, size=num_sensors, replace=False)
    sensor_positions = np.sort(sensor_positions)
    
    sensor_heights = []
    for pos in sensor_positions:
        idx = np.argmin(np.abs(x_array - pos))
        true_height = height_array[idx]
        
        if sensor_type.lower() == "lidar":
            noise_std = 0.1
        else:
            noise_std = 0.2
        
        measured = true_height + np.random.randn() * noise_std
        sensor_heights.append(max(measured, 0.0))
    
    return sensor_positions, np.array(sensor_heights)

def estimate_peak_height(sensor_positions, sensor_heights):
    """
    Simple approach: take the max sensor reading as approximate peak.
    """
    return np.max(sensor_heights)

def estimate_volume(x_array, height_array, silo_width=5.0, silo_depth=5.0):
    """
    Approximates volume of the pile using trapezoidal integration
    across the 2D cross-section, then multiplying by depth.
    
    Args:
        x_array: normalized x from 0..1
        height_array: height at each x
        silo_width (float): actual width of silo in meters
        silo_depth (float): depth into the page in meters
        
    Returns:
        volume (float): approximate volume in cubic meters
    """
    x_meters = x_array * silo_width
    area_2d = np.trapz(height_array, x_meters)  # m^2
    volume = area_2d * silo_depth               # m^3
    return volume

def plot_pile_shape(x_array, height_array, sensor_positions=None, sensor_heights=None):
    """
    Creates and returns a matplotlib figure showing
    the random pile shape and optionally sensor readings.
    """
    fig, ax = plt.subplots(figsize=(5,3))
    ax.plot(x_array, height_array, label='Pile Shape', linewidth=2, color='blue')
    
    if sensor_positions is not None and sensor_heights is not None:
        ax.scatter(sensor_positions, sensor_heights, color='red', 
                   label='Sensor Readings', s=50, zorder=5)
    
    ax.set_title("Silo Cross-Section (2D) - Wood Chip Pile")
    ax.set_xlabel("Normalized Silo Width (0..1)")
    ax.set_ylabel("Height (m)")
    ax.legend()
    fig.tight_layout()
    return fig

def plot_sensor_bar(sensor_positions, sensor_heights):
    """
    Creates and returns a bar chart showing sensor heights
    vs. sensor ID or position.
    """
    fig, ax = plt.subplots(figsize=(4,3))
    # Let's use sensor index as x-ticks
    indices = np.arange(len(sensor_positions))
    ax.bar(indices, sensor_heights, color='green')
    ax.set_title("Sensor Readings Bar Chart")
    ax.set_xlabel("Sensor Index")
    ax.set_ylabel("Measured Height (m)")
    ax.set_xticks(indices)
    ax.set_xticklabels([f"{pos:.2f}" for pos in sensor_positions], rotation=45)
    fig.tight_layout()
    return fig

# --------------------------------------
# Streamlit App
# --------------------------------------

st.set_page_config(page_title="ITC PSPD - Wood Chip Silo Demo", layout="centered")

st.title("Height Detection Of Wood Chips in Closed Silos - ITC PSPD Demo")

st.markdown("""
Welcome to a simple demonstration of *Wood-chip silo height profile Detection and Volume Estimation*.  
This is an **Industry 4.0** inspired solution to **digitally** track the *height* and *volume*  
of wood chips in a silo, supporting **paperboard and specialty paper** operations at **ITC PSPD**.

---
""")

# Sidebar controls
st.sidebar.header("Simulation Controls")

sensor_type = st.sidebar.selectbox("Select Sensor Type:", ["LiDAR", "Ultrasonic"])
max_height = st.sidebar.slider("Max Potential Height (m):", 5.0, 30.0, 10.0, 1.0)
noise_level = st.sidebar.slider("Random Noise Level:", 0.0, 3.0, 0.5, 0.1)
num_points = st.sidebar.slider("Pile Resolution (Points):", 20, 200, 50, 10)
num_sensors = st.sidebar.slider("Number of Sensors:", 1, 12, 5)
silo_width = st.sidebar.slider("Silo Width (m):", 1.0, 20.0, 5.0, 1.0)
silo_depth = st.sidebar.slider("Silo Depth (m):", 1.0, 20.0, 5.0, 1.0)

# Button to generate new random pile
if "x_array" not in st.session_state or "height_array" not in st.session_state:
    st.session_state["x_array"] = None
    st.session_state["height_array"] = None

if st.sidebar.button("Generate New Random Pile"):
    x_array, height_array = generate_random_pile(
        num_points=num_points,
        max_height=max_height,
        noise=noise_level
    )
    st.session_state["x_array"] = x_array
    st.session_state["height_array"] = height_array

# If we have data, proceed
if st.session_state["x_array"] is not None and st.session_state["height_array"] is not None:
    x_array = st.session_state["x_array"]
    height_array = st.session_state["height_array"]
    
    # Simulate sensor readings
    sensor_positions, sensor_heights = simulate_sensor_readings(
        x_array, height_array, sensor_type, num_sensors
    )
    
    # Estimate peak height from sensor
    approx_peak = estimate_peak_height(sensor_positions, sensor_heights)
    # Estimate volume
    volume_est = estimate_volume(x_array, height_array, silo_width, silo_depth)
    
    st.subheader("1) Silo Cross-Section Visualization")
    fig_pile = plot_pile_shape(x_array, height_array, sensor_positions, sensor_heights)
    st.pyplot(fig_pile)
    
    # Show measured peak height
    col1, col2 = st.columns(2)
    with col1:
        st.markdown(f"**Approx Peak Height (from sensors):** `{approx_peak:.2f} m`")
        st.progress(min(approx_peak/max_height, 1.0))
    with col2:
        st.markdown(f"**Estimated Volume:** `{volume_est:.2f} m³`")
        st.progress(min(volume_est/((max_height*silo_width*silo_depth)), 1.0))
    
    st.subheader("2) Sensor Data Overview")
    
    col3, col4 = st.columns(2)
    with col3:
        # Show bar chart of sensor readings
        fig_bars = plot_sensor_bar(sensor_positions, sensor_heights)
        st.pyplot(fig_bars)
    with col4:
        st.write("### Sensor Data Table")
        table_data = {
            "Position (norm)": [f"{pos:.2f}" for pos in sensor_positions],
            "Height (m)": [f"{h:.2f}" for h in sensor_heights]
        }
        st.table(table_data)
    
    st.markdown("---")
    
    st.markdown("""
    ### Interpretation & Industry 4.0 Connection

    - **Digital Twins**: This simulation mimics how a *digital twin* of the silo tracks
      wood-chip heights. The real system could have *LiDAR* sensors scanning
      or an array of *ultrasonic* sensors measuring the pile at intervals.
    - **Data-Driven Insights**: Automated volume estimation helps production planning, 
      ensuring continuous supply for **paperboard** manufacturing while minimizing 
      waste or silo overflow.
    - **ITC PSPD**: As a leading division in paper & packaging, adopting *Industry 4.0* 
      solutions supports sustainability and operational efficiency—aligned with ITC’s 
      triple bottom line.
    - **Big Data & AI**: If multiple silos or sites feed data into a cloud platform, 
      advanced analytics/predictive models can optimize inventory or detect anomalies.

    ---
    """)
else:
    st.info("Use the **sidebar** to generate a new random pile and view the results.")

st.markdown("""
## Made By:
- Name: Kaustubh Raykar 
- PRN: 21070126048
- Btech AIML 2021-25
- Contact: +91 7020524609
- Symbiosis Institute Of Technology, Pune 
- [email protected]
- [email protected]

**Thank you** for exploring this Wood Chip Silo Demo for **ITC PSPD**.
""")