Google ADK Multimodal Flowmeter Diagnostics Agent with Gemini
Industrial flowmeter calibration drifts cause costly billing errors and mass balance discrepancies in process plants. Construct a multimodal diagnostic agent using Google ADK and Gemini to parse acoustic sensor signals, camera readings, and flow rate logs. The agent must accurately identify 95% of calibration anomalies and issue corrective valve actions.
What you are building
Develop a Google ADK multimodal agent leveraging Gemini 1.5 Pro to analyze flowmeter acoustic waveforms, visual display photos, and time-series telemetry.
How work is evaluated
Evaluates multimodal flowmeter anomaly classification accuracy and calculated calibration drift offset validity.
Datasets
Public datasets, plus the private datasets you have access to.
How submissions are scored
multimodal_fault_classification
Checks if aerated liquid flow failure is correctly distinguished from physical fouling
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
diagnostic_accuracy
Accuracy of anomaly identification across test set • target: 0.95 • range: 0-1
This dimension contributes its full weight only when the submission satisfies the requirement. Partial credit is not awarded.
Learning goals
Build a multimodal agent using Google Agent Development Kit (ADK) and Gemini
Process visual images of flowmeter physical dials alongside acoustic vibration audio spectrographs
Correlate sensor physical state with time-series flow rate log discrepancies
Automate generation of zero-drift calibration offset parameters
Resources
Multimodal dataset containing flowmeter display images, acoustic spectrographs, and raw time-series flow logs across clean and fouled operational states.
Challenge setup
Evaluates multimodal flowmeter anomaly classification accuracy and calculated calibration drift offset validity.
Challenge input
JSON containing image_url, audio_spectrograph_url, and telemetry_log array
Google ADK
Assigned framework for building agents with Google Vertex/Gemini tools.
Evaluated output
JSON with fault_identified (string), drift_percentage (float), recommended_recalibration_offset (float)
- Checks if aerated liquid flow failure is correctly distinguished from physical fouling
- Accuracy of anomaly identification across test set • target: 0.95 • range: 0-1
- Benchmark: Multimodal Industrial Diagnostics Benchmark
- Diagnostic Accuracy target: 0.95
- 1 public reference case
- Python execution harness
- Python sandbox (unavailable on Versalist)
Shows the first task, up to three tools, and up to three evidence items.
View technical recipe
Configured tools
- Google · Required
- Google ADK · Optional
- Google Jules · Optional
- Google Jules · Optional
Evaluation contract
- multimodal_fault_classification · Weight 1
- diagnostic_accuracy · Weight 1
Metrics
- diagnostic_accuracy · Target 0.95
Benchmarks
- Multimodal Industrial Diagnostics Benchmark
- robotics
Reference cases
- 1 public reference case
Workflow tasks
- Task 1
Input: JSON containing image_url, audio_spectrograph_url, and telemetry_log array
Output: JSON with fault_identified (string), drift_percentage (float), recommended_recalibration_offset (float)
Execution harness: Python
Evaluation tools
- Google ADK · Multimodal Agent Framework
Recipe freeze marker
No recipe freeze time is recorded.