ETAS Calibration Suite

Welcome to ETAS Calibration Suite

AI-powered end-to-end calibration workflow

Cal Suite

Your AI concierge can guide you through the entire calibration pipeline. Use the quick actions or ask anything.

New Project

Set up a new calibration project

Open project

Find and resume a recent project

Standalone Tools

Use independently without the full calibration workflow

EATB

Data visualization & reporting

localhost:9231

Data Operator

Parse, curate & transform data

Currently unavailable

ASCMO-MOCA

Model calibration & optimization

Currently unavailable

AI Active Learning Loop: Observe → Learn → Recommend → Execute Live

ETAS Engineering Sidekick

Step-by-step guidance • 16 knowledge agents • Calibration process

ASCMO EATB FMU +10

Welcome! I'm your ETAS Engineering Sidekick. I can:

Guide you step-by-step through any workflow
Answer questions from the ETAS knowledge base
Help plan calibration (DPF, EGR, T2.0, etc.)
Write automation scripts

What would you like to do?

Workflow guide T2.0 Cal FMU help Sigma plan

AI Smart Recommendations

Proactive insights based on your project state and data patterns

Active Learning NTR Engine

1 Observe → 2 Learn → 3 Recommend → 4 Execute → ↺ continuous

Project Setup

Define optimization scope, targets, and constraints — Lead Calibrator

Step 2

Model Upload

Upload your model for FMU generation

Step 2

Upload your model for FMU generation or validation. Supported: C source code (.c, .h), project archives (.zip with sources), pre-built FMU files (.fmu), and static libraries (.a, .lib). For Simulink or MATLAB models, first export as C-code using Embedded Coder, then upload the generated .c/.h files here.

Drop model files here or click to browse

Supports: .c, .h, .zip, .fmu, .a, .lib (max 500MB)

Model Fidelity Check

AI-powered validation against ETAS knowledge base

Not run

The AI fidelity engine validates your model against ETAS lessons learned and best practices. It continuously learns from new patterns and curated feedback.

FMU Generation

Build FMUs with ASCMO FMU Builder

Step 4

Upload your source files and select target platform(s). The system will compile your model into FMI 2.0 compliant FMUs.

About the FMI Wrapper: If you only have your model C-code (e.g., Turbo_Boost_Model.c), the tool will auto-generate the FMI 2.0 wrapper (_FMU.c) for you — including fmi2Instantiate, fmi2DoStep, fmi2GetReal/SetReal, and correct causality classification.

If you already have an FMI wrapper file (a _FMU.c that includes fmi2Functions.h), upload it alongside your model code and the tool will use yours instead.

How to Export C-Code from Your Model

Detailed instructions for exporting from Simulink (with/without license), OpenModelica, Scilab, Python, ASCET, GT-SUITE, and more.

Open Guide ↗

Source Files

Drop C source files here (.c, .h)

Accepts .zip ECU projects, C source files (.c, .h), static libraries (.a, .lib), or Embedded Coder exports

Target Platform(s)

Windows + Linux

Generate both platform FMUs

Linux Only

Generate Linux FMU only

ASCMO-MOCA Project Setup

Create your MOCA optimization project on the desktop tool, then upload here

Step 6

Already have a .moca project file? Skip this step and go directly to Data Acquisition →

How to Create a MOCA Project in ASCMO Desktop

1

Open ASCMO-DESK → Model Calibration (MOCA)

Launch the ASCMO-DESK application and click the "Model Calibration" tile to open ASCMO-MOCA.

2

Create New Project → Import FMU

Click "New" in the start window. In the main workspace, click "Import FMU" and select your .fmu file (built in Step 5 or pre-existing).

3

Define Parameters → Create or Load Variables

In the Parameters step, create or import your FMU variables. MOCA reads the variable list from the FMU's modelDescription.xml. You then assign roles: Inputs (factors to vary), Outputs (responses to optimize), Parameters (constants to tune). This must be done before mapping.

4

Define Objectives & Constraints

Set optimization objectives (minimize NOx, minimize BSFC, etc.) and hard constraints (OBD limits, temperature bounds). These should match what you defined in Project Setup (Step 2).

5

Configure Measurement Data Import

Import your measurement data files. MOCA uses this data to train the model internally. Ensure the data channels match the variable names defined in Step 3. MOCA handles its own internal training data allocation.

6

Save Project → Export for MOCA-Runtime

File → Save (as .moca). Optionally: File → Export to MOCA-Runtime for cloud execution. The saved .moca file is what you'll upload in Step 9 (MOCA Optimization).

Key Requirements for Cloud Optimization

• Causality must be correct: inp_* → Input, out_* → Output, par_* → Parameter

• FMU must contain a step function (calculate(dt) or similar) for time-stepping

• Same FMU used in desktop MOCA and uploaded to cloud — no version mismatch

• Objectives in MOCA should match the targets defined in Project Setup (Step 2)

Detailed MOCA Guide (for novice users — click to expand)

Understanding Gaussian Process Models

MOCA uses Gaussian Process (GP/Kriging) regression — a non-parametric modeling approach that provides both a prediction AND an uncertainty estimate at every point. Key properties:

Interpolating: Passes exactly through training data points
Uncertainty quantification: Shows wider confidence bands far from data
Automatic Relevance Determination (ARD): Identifies which inputs matter most
Kernel options: Squared Exponential (RBF), Matérn 3/2, Matérn 5/2

Target accuracy: R² > 0.95 for good optimization results. If R² < 0.90, add more data points.

Design of Experiments (DoE) — Recommended Sizes

Factors	Min Points	Recommended	DoE Type
2-3	30	40-50	Latin Hypercube
4-5	50	60-80	Latin Hypercube
6-8	80	100-150	D-optimal or LHS
9+	120+	150-200+	Space-filling + Adaptive

Rule of thumb: Minimum = 10 × number_of_factors. Add 15-25 adaptive points after initial model for improved accuracy at boundaries.

Multi-Objective Optimization (Pareto Front)

When objectives conflict (fuel vs NOx), MOCA finds the Pareto front — the set of solutions where no objective can be improved without worsening another.

Algorithm: NSGA-II (Non-dominated Sorting Genetic Algorithm)
Population size: 100-500 (larger = better coverage, slower)
Generations: 200-1000 (more = finer Pareto resolution)
You select your preferred trade-off point from the Pareto curve

FMU Virtual Calibration Workflow

Instead of running physical tests, MOCA can optimize using an FMU (simulation model):

Load FMU (your plant model from Step 5)
Define FMU inputs as: operating conditions + calibration parameters
Define FMU outputs as: responses to optimize (NOx, BSFC, etc.)
Run space-filling DoE on FMU (automated, no test bench needed)
Train GP surrogate models from FMU responses
Optimize on GP surrogates (1000× faster than running FMU)
Validate optimal points by running them back through FMU
Export optimized calibration maps (DCM/CDFX)

Map Generation & Export

Convert scattered optimized points → regular grid calibration map
Apply smoothing (gradient constraints for drivability)
Export as: DCM (INCA format), CDFX (A2L compatible), or CSV

Common Troubleshooting

Problem	Cause	Solution
R² < 0.90	Not enough data / poor coverage	Add 20+ adaptive DoE points at high-uncertainty regions
Optimization finds infeasible solutions	Constraint boundaries too tight	Add margin to constraints (e.g., NOx < 55 instead of < 60)
Maps not smooth	Smoothing weight too low	Increase smoothing parameter in map generation (0.3-0.7)
Very long training time	Too many data points (>5000)	Use subset selection or sparse GP approximation
Pareto front is flat	Objectives not truly conflicting	Check if single-objective optimization is sufficient

Data Pre-Staging

Prepare and filter measurement data using Data Operator before cloud upload

Step 7

Before uploading data to the cloud, use ETAS Data Operator on your desktop to filter, clean, and reduce your measurement files. This saves cloud storage costs and ensures only relevant data enters the pipeline. The AI Sidekick can help you identify which channels to keep.

Why Pre-Stage?

90%

Typical data reduction

~$0

No cloud cost for filtering

Faster

Smaller uploads & queries

Pre-Staging Process Flow

Raw MF4
10 files, 500+ ch

→

AI Scan
Recommend channels

→

DO Filter
Select & process

→

Clean Export
1 file, 25-40 ch

→

Upload
Step 8

Supported Measurement Formats

.mf4

ASAM MDF4

.dat/.mdf

Legacy INCA

.csv

CSV / TSV

.mat

MATLAB v5/7.3

.xlsx

Excel

.tdms

NI DIAdem

.dcm

Calibration

.asc

CAN trace

Step-by-Step: Data Operator Pre-Staging Guide

1

Launch Data Operator

Start Menu → ETAS → Data Operator
Or PowerShell: & "C:\Program Files\ETAS\DataOperator\DataOperator.exe"

Keyboard Shortcut

If pinned to taskbar: Win + number key

2

Import Measurement Files

File → Import or drag files directly into the workspace.
Select all your MF4/MDF files. Data Operator will merge them into a unified view.

Import Dialog Settings:

Merge common channels

Auto-detect time base

Preserve units & metadata

Do NOT resample on import

3

Ask AI: Which channels do I need?

The AI Sidekick knows your project context (from Step 2) and the Data Operator channel naming conventions. Ask it to recommend relevant channels.

Which channels do I need? Steady-state settings? Filter settings for temps?

4

Select & Filter Channels

In the Channel List panel, select only the AI-recommended channels.

Data Operator Channel Selection:

1. Channel List → Right-click → "Select All" → then Deselect All
2. Use the filter bar at the top to search (e.g., type "temp" to find temperature signals)
3. Ctrl+click to select multiple channels
4. Right-click selected → "Mark for Export"

Tip: Always include these base channels: Engine Speed (RPM), Torque/Load, Coolant Temp, Ambient Temp, and your time channel. These are needed by almost every calibration model.

5

Apply Data Processing (Optional but Recommended)

Use Data Operator's built-in tools to clean and condition your signals before export.

Steady-State Detection

Menu: Tools → Steady-State Detection
Speed tolerance: ±10-20 rpm
Torque tolerance: ±2-5%
Temp gradient: < 1-2°C/min
Min window: 10-30 seconds
Output: Extracts mean values at each stable point

Low-Pass Filtering

Menu: Tools → Filter → Low-Pass
Temperature signals: 0.5-1 Hz cutoff
Pressure signals: 5-10 Hz cutoff
Emissions (NOx, PM): 1-2 Hz cutoff
Type: Butterworth, 2nd-4th order
Removes sensor noise without losing dynamics

Resampling

Menu: Tools → Resample
ASCMO-STATIC: Not needed (use SS points)
ASCMO-DYNAMIC: 10 Hz common base
Combustion: Keep 1000 Hz
Method: Linear interpolation
Aligns signals to same time base

Outlier Removal

Menu: Tools → Outlier Detection
Z-score: Remove > 3σ from mean
IQR: Outside Q1-1.5×IQR to Q3+1.5×IQR
Spike filter: Median filter, window=5
Prevents bad data from corrupting models

Common Derived Calculations

                    BMEP = (Torque × 4π) / (Displacement × 2)

                    Power_kW = Torque × RPM × (2π/60) / 1000

                    Lambda = AFR / 14.7

                    Vol_Eff = (MAF × 2) / (Disp × ρ_air × RPM/60)

6

Export Reduced Dataset

Export only the selected and processed channels to a new file.

Export Settings:

Menu: File → Export → Selected Channels
Format: MF4 (recommended) or CSV
Naming: [project]_[date]_filtered.mf4
Content: Only marked channels + derived signals

MF4 Export

Preserves units & metadata

CSV Export

Universal compatibility

Ready for Upload

Your filtered file is ready. Continue to Step 8 (Data Upload) where the clean data enters the Bronze → Silver → Gold pipeline. Only relevant channels will be indexed for AI queries and optimization.

Which Workflow? ASCMO-STATIC vs ASCMO-DYNAMIC

ASCMO-STATIC (Steady-State)

Use when: Map calibration, DoE at operating points
Data Operator steps:
1. Import raw MF4
2. Filter noise (LP filter)
3. Run steady-state detection
4. Extract mean values per OP
5. Export as CSV (1 row per point)

ASCMO-DYNAMIC (Time-Series)

Use when: Transient model, system ID, NARX
Data Operator steps:
1. Import CAN/MDF recording
2. Resample to 10 Hz common base
3. Filter sensor noise
4. Segment into maneuvers
5. Export time-series segments

Batch Processing: If you have many files with the same channel structure, use Data Operator's Batch Mode (Tools → Batch Processing) to apply the same filter + export pipeline to all files automatically. Define your pipeline once, run on a folder.

Alternative: Cloud-Side AI Filter

Don't have Data Operator installed? In the next step (Data Upload), the cloud offers an AI Smart Filter that scans your file headers and lets you select channels before ingestion. It costs ~1K tokens but requires uploading the full file temporarily for scanning.

Desktop pre-staging = free, faster, more processing options (filtering, resampling, steady-state detection). Cloud filter = convenient, small token cost, no desktop tool needed.

Already have clean data? If your files are already filtered (single-channel CSV from test bench, or previously processed MF4), skip this step and go directly to upload.

Data Acquisition

Upload or pull measurement data

Step 8

Additional measurements are optional. Upload them from your PC or pull from a connected cloud server when needed. Uploaded data is processed through the Medallion architecture (Bronze → Silver → Gold) before being made available to the LLM and optimization tools.

Upload Data from PC

Drop measurement files here or click to browse

Supports: .mf4, .mdf, .csv, .xls, .xlsx

Coming later: .c, .h, .mat, .dat, .dcm, .txt, .tdms, .asc, .tsv, .mdf4

EATB Pre-Optimization Visualization

Visualize raw data before optimization

Step 9

Visualization Tool

Token Usage Note: Python-based analytics require LLM-generated scripts, consuming more tokens per session. Using EATB or Data Operator through your license executes natively without additional LLM calls, significantly reducing token/credit consumption.

EATB (Web)

Interactive test bench dashboard

Licensed ✓

Lower token usage

Python (Plotly)

Auto-generated Python dashboard

Always Available

Higher token usage

Custom Plot Builder — Your Data

Select signals from your uploaded measurement data to create interactive charts. All plots are generated from your actual data.

Upload data in Step 8 to see available signals, or select from detected channels:

X-Axis

Y-Axis (hold Ctrl for multi-select)

Chart Type

EATB Dashboard Setup Guide (for novice users — click to expand)

What is EATB?

ETAS Automation Toolbox (EATB) is a web-based visualization and automation platform for calibration data. It runs locally on port 9231 and provides interactive dashboards, reports, and scripting capabilities.

Setting Up EATB on Your PC

Install: EATB is included with ETAS Tool Suite. Verify: Start Menu → ETAS → EATB
Configuration: Edit eatbconfiguration.properties to set port (default: 9231)
Start: Run EATB service — it will serve dashboards at http://localhost:9231
Verify: Open browser → http://localhost:9231 → you should see the EATB home page

EATB Dashboard Features

Time-Series Viewer

Plot any signal vs time. Zoom, pan, compare multiple channels. Cursor readout at any point.

Correlation / Scatter

X vs Y scatter plots. Color by 3rd variable. Trend lines, R² display.

Histograms & Distributions

Data distribution analysis. Bin count adjustment. Normal curve overlay.

Traffic Light Reports

Pass/fail status per operating point. Green/yellow/red based on tolerance bands.

Map Viewer (3D)

3D surface plots of calibration maps. Pre vs post comparison. Difference maps.

Campaign Summaries

KPI rollups per test campaign. Automated progress reporting.

EATB Python Scripting (Automation)

# Connect to EATB

import eatb

session = eatb.connect('localhost', 9231)

# Load measurement data

data = session.load_file('measurement.mf4')

# Create time-series plot

plot = session.create_plot('TimeSeries')

plot.add_signal('EngineSpeed')

plot.add_signal('ExhaustTemp_Bank1')

plot.show()

# Generate traffic light report

report = session.create_report('TrafficLight')

report.add_criterion('NOx', max=60, unit='mg/km')

report.add_criterion('Soot', max=4.5, unit='mg/km')

report.generate()

Demo Templates

The inline dashboard above uses Plotly.js charts that mirror EATB's visualization style. Available chart types:

Line (Time Series): Select time on X-axis, any signal(s) on Y-axis
Scatter (X vs Y): Any two signals as X and Y (e.g., RPM vs NOx)
Histogram: Distribution of any single signal
Box Plot: Statistical summary across operating conditions

To use with your data: Upload measurement files in Step 8 (Data Upload). The Custom Plot Builder above will auto-populate with your actual signal names.

ASCMO-MOCA Optimization

Multi-objective calibration optimization

Ready

How ASCMO-MOCA Cloud Optimization Works

Simply upload your .fmu and .moca project file. The platform handles everything else automatically:

You upload: *.fmu *.moca (project) → AI auto-generates: script.py work_package.json orchestration.json

Fully automated: Once you upload the two files, the AI generates all execution scripts, validates them, assembles the package, and submits to the cloud orchestrator. No manual scripting required.

①

Upload

Drop your .fmu + .moca files

②

AI Generates & Validates

Scripts, configs, orchestration payload — all auto-generated and syntax-checked

③

Select Compute & Run

Pick your tier, click Start — cloud handles the rest

Upload Optimization Inputs

Provide an .fmu, .moca project, and required .ascmoScript. Any Step 8 measurements are included automatically.

1. FMU File

Drop .fmu here

Your compiled FMU model

2. MOCA Project File

Drop .moca here

ASCMO-MOCA project config

3. ASCMO Script

Drop .ascmoScript here

Required ASCMO execution script

4. Additional measurements (optional)

Select Compute Tier

Azure D-series v5 VMs (US East 2) — on-demand pricing. Time estimates adjust dynamically based on uploaded FMU size and data volume.

16

vCPUs

D16ds v5 • 64 GB

$0.768/hr

~4.2h • $3.23/run

32

vCPUs

D32ds v5 • 128 GB

$1.536/hr

~2.3h • $3.53/run

64

vCPUs

D64ds v5 • 256 GB

$3.072/hr

~1.4h • $4.30/run

Best value

128

vCPUs

2× D64ds v5 • 512 GB

$6.144/hr

~0.9h • $5.53/run

Fastest

Speed vs. Cost Efficiency

Dynamic estimation based on uploaded file sizes. Bar chart shows time per tier; line shows total cost. Green = recommended.

I acknowledge that the data being processed satisfies data privacy requirements and is authorized for cloud processing.

Demo mode — simulated execution. Sign in to run ASCMO optimization.

Post-Optimization Results & Reporting

Compare pre/post optimization and generate reports

Complete

View Results Using:

EATB (Web)

Pre vs Post comparison dashboard

Licensed ✓

Lower token usage

Python Analytics

Full statistical comparison with heatmaps, scatter, FFT

Always Available

Higher token usage

—

Model Quality (R²)

—

Fuel Economy

—

Pareto Solutions

—

OBD Compliance

AI Post-Optimization Analysis Auto-generated

Optimization Summary

Risks & Follow-up

Report Generation

Usage & Billing Dashboard

Token consumption, costs, transactions, and credit tracking

Pay-Per-Use

Tokens Used (This Month)

—

Loading...

ROI Value (Time Saved)

—

Loading...

Compute Cost (MTD)

—

Loading...

Total Cost (MTD)

—

Loading...

Daily Token Consumption (Real-time)

Cost Breakdown (Live)

Usage by Team Member (Live)

ROI: Hours Saved by Category

Recent Transactions

Timestamp	User	Action	Tokens	Cost	Agent
Jul 15, 14:32	suresh.s	Chat query	2,847	$0.043	ASCMO-MOCA
Jul 15, 14:28	jerry.c	FMU Build (8-core)	—	$3.20	FMU Builder
Jul 15, 14:15	suresh.s	Python analytics	8,421	$0.126	Python
Jul 15, 13:52	jerry.c	MOCA optimization	—	$12.80	ASCMO-MOCA
Jul 15, 13:30	suresh.s	Data ingest (50MB)	1,205	$0.018	Data Operator

Infrastructure & Platform Services

Cross-cutting services, compute, storage, compliance, and monitoring

All Systems Operational

Cross-Cutting Platform Services

Microsoft Marketplace

Billing & Transactions

Active

Cloud Licensing & Metering

Usage tracking, entitlements

Active

Security & Compliance

GDPR, PII, Data Governance

Compliant

Notifications & Alerts

Email / In-App

Active

Audit & Compliance

Access, Actions, Traceability

Logging

Cost Management

CPU Allocation, Estimation

Tracking

Infrastructure & Core Platform

Secure Cloud Environment

Tenant Isolation, IAM, Encryption

● Azure East US — Healthy

Orchestration & Scheduling

Kubernetes, Autoscaling, Jobs

● 3 nodes — 2 jobs queued

Compute Allocation

CPU/Memory, Cost & Time Based

● 8-core tier — 62% utilized

Data Lake Storage

Secure, Scalable, High Availability

● 2.4 TB used / 10 TB quota

Model & Artifact Storage

Reports, DCM, Models, Logs

● 847 artifacts — 12.3 GB

Monitoring & Observability

Metrics, Logs, Traces, Health

● 99.97% uptime (30d)

ETAS Cloud Hub License Status

EATB

Licensed ✓

Data Operator

Licensed ✓

ASCMO-MOCA

Licensed ✓

FMU Builder

Licensed ✓

Lessons Learned & Best Practices

Your organization's collective calibration knowledge

—

0

Total Conversations

0

Lessons Learned

0

AI Corrections

0

Team Members

Add Lesson Learned / Best Practice

Company Knowledge Graph

Tenant-Isolated

Upload your organization's proprietary knowledge — calibration guides, internal standards, process documents, model libraries. This knowledge is only accessible to users from your company and is used by the AI to provide company-specific recommendations.

Knowledge domains Leave empty for automatic routing; select multiple domains to override.

Upload Documents

PDF, DOCX, TXT, MD, CSV

Import Knowledge DB

JSON or SQLite export

Recent Lessons & Best Practices

Save & Archive

Save project permanently and manage data retention

Step 12

Project workflow state and final reports remain available according to your customer retention policy. File expiry is calculated by the server for this project.

Permanent Storage

Project configuration (Step 2)

Optimization results & Pareto selections

Final reports & exported calibrations (DCM/CDFX)

AI data insights & recommendations

Annotations, audit trail, version history

Model health validation records

Retained indefinitely — always accessible

Project File Retention

Retention status loads when the project is resumed.

Expired projects permit project details and final report viewing only.

The external retention job removes expired files.

Resume Project

Resume restores saved steps, current execution state, and linked files while the project is active. After expiry, the project becomes read-only and files are unavailable.

What Will Be Saved

Project: —

FMU: No FMU built yet

MOCA Project: No MOCA project uploaded

Additional measurements: No additional measurements

Results: No optimization results

AI Insights: Chat history & recommendations saved

ETAS Engineering Sidekick

Natural language interface to all calibration tools

LLM Connected MCP Active

Signal-ID	Unit	Min	Max	Mean	Status
EngSpd	RPM	680	4200	1850	●OK
T2_ExhTemp	°C	142	687	385	●OK
Lambda	−	0.88	1.42	1.01	●Warn

Metric	Pre-Opt	Post-Opt	Δ Change	Status
NOx (ppm avg)	142.3	125.2	-12.0%	●Pass
BSFC (g/kWh)	238.1	230.5	-3.2%	●Pass
Soot (mg/m³)	18.4	16.9	-8.2%	●Pass
T2.0 Peak (°C)	692	678	-2.0%	●Pass

Configure Project Setup Page

Understanding Gaussian Process Models

Design of Experiments (DoE) — Recommended Sizes

Multi-Objective Optimization (Pareto Front)

FMU Virtual Calibration Workflow

Map Generation & Export

Common Troubleshooting

Measurement preview

What is EATB?

Setting Up EATB on Your PC

EATB Dashboard Features

EATB Python Scripting (Automation)

Demo Templates

Run log