Skip to content
Tier0 Docs

Analyze UNS Data

Available inCloudEnterprise

Tier0 adopts Marimo Notebook for advanced data analysis with Python.

Creating Notebooks

Section titled “Creating Notebooks”
  1. Log in to Tier0, and select Notebook.
  2. Hover over the +, click New Notebook.
  3. Enter the notebook Name and click Create.
  4. Click and enter the created notebook to start editing.

Querying Data from UNS

Section titled “Querying Data from UNS”

  1. Inside the notebook, click the VARIABLES on the left side.

  2. Expand the PostgreSQL list, select a table under tenant_id > uns.

  3. Hover over the table, and click the + icon next to the table name to generate SQL statements cell automatically.

    Terminal window
    SELECT * FROM "uns"."table_name" LIMIT 100

  4. Click the Run icon at the lower-right corner to run all cells and get the query result.

Analyzing UNS Data

Section titled “Analyzing UNS Data”

Example Background

Section titled “Example Background”
  • Data source is temperature, pressure, status and vibration of a compressor collected from a factory and sent to Tier0 UNS.
  • Isolation Forest is used to recognize abnormal data points in the dataset.
  • Detection sensitivity can be adjusted, and the results are visualized accordingly in a dashboard.

Operations

Section titled “Operations”
  1. Create a notebook and access it.
  2. Add the following cells in order.
Cell 1: Import dependencies.
Terminal window
import marimo as mo
import pandas as pd
import numpy as np
import altair as alt


from sklearn.preprocessing import StandardScaler
from sklearn.ensemble import IsolationForest
Cell 2: Get data from UNS, clean it to get the analysis dataset.
Terminal window
status_pd = status_data.to_pandas()
temperature_pd = temperature_data.to_pandas()
pressure_pd = pressure_data.to_pandas()
vibration_pd = vibration_data.to_pandas()




# Convert and sort timestamps
for frame in [
    status_pd,
    temperature_pd,
    pressure_pd,
    vibration_pd,
]:
    frame["_timestamp"] = pd.to_datetime(
        frame["_timestamp"],
        utc=True,
        errors="coerce",
    )


    frame.dropna(
        subset=["_timestamp"],
        inplace=True,
    )


    frame.sort_values(
        "_timestamp",
        inplace=True,
    )




# Use temperature as the main timeline
sensor_data = pd.merge_asof(
    temperature_pd,
    pressure_pd,
    on="_timestamp",
    direction="nearest",
    tolerance=pd.Timedelta(seconds=30),
)




sensor_data = pd.merge_asof(
    sensor_data,
    vibration_pd,
    on="_timestamp",
    direction="nearest",
    tolerance=pd.Timedelta(seconds=30),
)




sensor_data = pd.merge_asof(
    sensor_data,
    status_pd,
    on="_timestamp",
    direction="nearest",
    tolerance=pd.Timedelta(seconds=30),
)




# Remove incomplete records
complete_data = sensor_data.dropna(
    subset=[
        "temperature",
        "pressure",
        "vibration",
        "status_code",
    ]
)




# Analyze only active running periods
running_data = (
    complete_data[
        complete_data["status_code"] == 1
    ]
    .reset_index(drop=True)
)




running_data
Cell 3: Set the slider for anomaly detection threshold.
Terminal window
anomaly_threshold = mo.ui.slider(
    start=-0.15,
    stop=0.10,
    step=0.01,
    value=0.00,
    label="Anomaly detection threshold",
)




mo.vstack([
    mo.md("## Model Settings"),
    mo.md(
        """
Adjust the anomaly detection threshold.


A higher threshold applies stricter criteria and identifies more operating records as anomalies.
        """
    ),
    anomaly_threshold,
])
Cell 4: Use the Isolation Forest model to detect anomalies.
Terminal window
feature_names = [
    "temperature",
    "pressure",
    "vibration",
]




# Prepare model input
model_input = running_data[
    feature_names
]




# Standardize sensor values
feature_scaler = StandardScaler()


scaled_features = feature_scaler.fit_transform(
    model_input
)




# Train the anomaly detection model
anomaly_model = IsolationForest(
    n_estimators=200,
    contamination="auto",
    random_state=42,
)




anomaly_model.fit(
    scaled_features
)




# Calculate continuous anomaly scores
anomaly_scores = anomaly_model.decision_function(
    scaled_features
)




# Prepare analysis results
ml_result = running_data.copy()


ml_result["anomaly_score"] = anomaly_scores




# Classify records using the interactive threshold
ml_result["condition"] = np.where(
    ml_result["anomaly_score"] < anomaly_threshold.value,
    "Anomaly",
    "Normal",
)




ml_result
Cell 5: Displays results on a dashboard.
Terminal window
# Calculate summary metrics
dashboard_total = len(ml_result)


dashboard_normal = int(
    (
        ml_result["condition"] == "Normal"
    ).sum()
)


dashboard_anomaly = int(
    (
        ml_result["condition"] == "Anomaly"
    ).sum()
)




dashboard_anomaly_ratio = (
    dashboard_anomaly / dashboard_total * 100
    if dashboard_total > 0
    else 0
)




# Calculate anomaly severity
anomaly_records = ml_result[
    ml_result["condition"] == "Anomaly"
]




if len(anomaly_records) > 0:
    average_anomaly_severity = float(
        (
            anomaly_threshold.value
            - anomaly_records["anomaly_score"]
        )
        .clip(lower=0)
        .mean()
    )
else:
    average_anomaly_severity = 0.0




# Illustrative equipment health score
severity_penalty = min(
    average_anomaly_severity * 80,
    20
)


dashboard_health_score = max(
    0,
    100
    - dashboard_anomaly_ratio * 1.2
    - severity_penalty
)




# Determine risk level
if dashboard_health_score < 60:
    dashboard_risk = "High"
    dashboard_conclusion = (
        "The compressor shows a high level of unusual operating behavior. "
        "Further equipment inspection is recommended."
    )


elif dashboard_health_score < 80:
    dashboard_risk = "Medium"
    dashboard_conclusion = (
        "Some unusual operating patterns were detected. "
        "Continue monitoring the equipment and review the abnormal periods."
    )


else:
    dashboard_risk = "Low"
    dashboard_conclusion = (
        "Most operating records follow the learned normal pattern. "
        "No significant abnormal behavior was detected."
    )




# Header
dashboard_header = mo.Html(
    """
    <div style="
        padding: 4px 0 18px 0;
        font-family: Arial, sans-serif;
    ">
        <div style="
            font-size: 26px;
            font-weight: 700;
            color: #1f2937;
            margin-bottom: 6px;
        ">
            Air Compressor Anomaly Detection
        </div>


        <div style="
            font-size: 14px;
            color: #6b7280;
            line-height: 1.5;
        ">
            Machine learning analysis based on temperature,
            pressure, and vibration during active running periods.
        </div>
    </div>
    """
)
# Conclusion card
dashboard_conclusion_card = mo.Html(
    f"""
    <div style="
        background: #f8fafc;
        border: 1px solid #e5e7eb;
        border-left: 4px solid #4f46e5;
        border-radius: 8px;
        padding: 16px 18px;
        margin-bottom: 18px;
        font-family: Arial, sans-serif;
    ">
        <div style="
            font-size: 12px;
            font-weight: 700;
            color: #6b7280;
            text-transform: uppercase;
            letter-spacing: 0.6px;
            margin-bottom: 6px;
        ">
            Analysis Summary
        </div>


        <div style="
            font-size: 15px;
            color: #1f2937;
            line-height: 1.6;
        ">
            {dashboard_conclusion}
        </div>
    </div>
    """
)
# KPI cards
dashboard_kpis = mo.Html(
    f"""
    <div style="
        display: grid;
        grid-template-columns: repeat(4, minmax(0, 1fr));
        gap: 14px;
        margin-bottom: 24px;
        font-family: Arial, sans-serif;
    ">


        <div style="
            border: 1px solid #e5e7eb;
            border-radius: 10px;
            padding: 16px 18px;
            background: white;
        ">
            <div style="
                font-size: 12px;
                color: #6b7280;
                margin-bottom: 8px;
            ">
                HEALTH SCORE
            </div>


            <div style="
                font-size: 28px;
                font-weight: 700;
                color: #111827;
            ">
                {dashboard_health_score:.1f}
                <span style="
                    font-size: 14px;
                    color: #9ca3af;
                ">
                    / 100
                </span>
            </div>
        </div>




        <div style="
            border: 1px solid #e5e7eb;
            border-radius: 10px;
            padding: 16px 18px;
            background: white;
        ">
            <div style="
                font-size: 12px;
                color: #6b7280;
                margin-bottom: 8px;
            ">
                RISK LEVEL
            </div>


            <div style="
                font-size: 28px;
                font-weight: 700;
                color: #111827;
            ">
                {dashboard_risk}
            </div>
        </div>




        <div style="
            border: 1px solid #e5e7eb;
            border-radius: 10px;
            padding: 16px 18px;
            background: white;
        ">
            <div style="
                font-size: 12px;
                color: #6b7280;
                margin-bottom: 8px;
            ">
                ANALYZED SAMPLES
            </div>


            <div style="
                font-size: 28px;
                font-weight: 700;
                color: #111827;
            ">
                {dashboard_total:,}
            </div>
        </div>




        <div style="
            border: 1px solid #e5e7eb;
            border-radius: 10px;
            padding: 16px 18px;
            background: white;
        ">
            <div style="
                font-size: 12px;
                color: #6b7280;
                margin-bottom: 8px;
            ">
                ANOMALY RATIO
            </div>


            <div style="
                font-size: 28px;
                font-weight: 700;
                color: #111827;
            ">
                {dashboard_anomaly_ratio:.1f}%
            </div>
        </div>


    </div>
    """
)




# Interactive anomaly chart
dashboard_chart = (
    alt.Chart(ml_result)
    .mark_circle(
        size=58,
        opacity=0.78,
    )
    .encode(
        x=alt.X(
            "_timestamp:T",
            title="Time",
        ),


        y=alt.Y(
            "vibration:Q",
            title="Vibration RMS (mm/s)",
        ),


        color=alt.Color(
            "condition:N",
            title=None,
            scale=alt.Scale(
                domain=[
                    "Normal",
                    "Anomaly",
                ],
                range=[
                    "#94a3b8",
                    "#ef4444",
                ],
            ),
        ),


        tooltip=[
            alt.Tooltip(
                "_timestamp:T",
                title="Time",
            ),


            alt.Tooltip(
                "temperature:Q",
                title="Temperature",
                format=".2f",
            ),


            alt.Tooltip(
                "pressure:Q",
                title="Pressure",
                format=".4f",
            ),


            alt.Tooltip(
                "vibration:Q",
                title="Vibration",
                format=".3f",
            ),


            alt.Tooltip(
                "anomaly_score:Q",
                title="Anomaly Score",
                format=".3f",
            ),


            alt.Tooltip(
                "condition:N",
                title="Condition",
            ),
        ],
    )
    .properties(
        width=980,
        height=380,


        title=alt.TitleParams(
            text="Detected Operating Anomalies",
            subtitle=(
                f"Threshold: {anomaly_threshold.value:.2f} · "
                f"{dashboard_anomaly:,} anomalies detected"
            ),
            anchor="start",
            fontSize=17,
            fontWeight=600,
            subtitleFontSize=12,
            subtitleColor="#6b7280",
            offset=14,
        ),
    )
    .configure_view(
        stroke=None
    )
    .interactive()
)




# Final dashboard
mo.vstack([
    dashboard_header,
    dashboard_conclusion_card,
    dashboard_kpis,
    mo.ui.altair_chart(
        dashboard_chart
    ),
])
  1. Click Run all stale cells at the lower-right corner to run all cells and view the results.

Next

Section titled “Next”