Extend Python AI for Additional Anomaly Detection

Job ID: 40178745

Budget: $750 – $1,500 USD

Title

Extend & Refactor Existing Python AI Diagnostic System
Offline Time-Series Analytics – Reuse Existing GitHub

Project Description

We have an existing Python-based diagnostic and AI analytics system on GitHub.
The current implementation successfully detects one anomaly type (e.g. valve leakage) using time-series machine data.

We are now looking for a developer to reuse, refactor, and extend this existing system to support additional anomaly types, while keeping the software offline, headless, and reusable.


Objective

The software will:

Process exported time-series data (batch / offline)

Reuse the existing analytics and AI logic

Extend detection to additional anomaly categories

Output simple, explainable diagnostic results for engineering use

The same analytics core will later be reused in other projects, so clean separation from UI and data acquisition is mandatory.

Budget

Single delivery, offline prototype

IMPORTANT – READ CAREFULLY

A working GitHub repository already exists and will be provided

This task includes reviewing, reusing, and extending existing code

It is NOT a rewrite

It is NOT open-ended AI research

Proposals assuming a rebuild from scratch will be rejected

Scope of Work
1. Review Existing Codebase

Review the current Python repository

Identify and reuse:

existing ingestion logic

feature extraction

AI / anomaly logic

Isolate or remove:

UI or demo code

hard-coded paths

non-essential scaffolding

2. Input Normalisation

Support exported time-series data files (CSV / XML or similar)

Map inputs into a common internal format, for example:

timestamp

channel / sensor type

value

unit

Batch / offline execution only

3. Feature & Anomaly Extension

Reuse existing feature calculations

Extend detection beyond valve leakage to additional anomaly types, such as:

abnormal vibration behaviour

instability or variability increase

pressure-related anomalies

trend-based degradation indicators

Anomalies may be implemented using:

simple ML extensions or

deterministic / rule-based logic where appropriate

Focus on engineering relevance and explainability, not model complexity


4. Offline AI / Analytics Execution

Extend existing AI logic to support multiple anomaly categories

Keep ML additions small and controlled

Provide rule-based fallback logic for clarity and robustness

5. Outputs

Generate simplified diagnostic outputs such as:

normal / warning / fault

anomaly category

basic confidence indicator

Output formats:

CSV

JSON

6. Execution

CLI or script-based execution, e.g.:

python run_analysis.py input_folder/ output_folder/


Must run locally on a standard laptop.

Deliverables

Updated Python code (Python 3.8+)

Clean folder structure

README explaining:

supported input formats

anomaly types implemented

how to run the analysis

Sample input data and example outputs

Explicitly Out of Scope


Required Skills

Python

Pandas / NumPy

Time-series analysis

Practical ML or anomaly detection

Experience refactoring existing codebases

Nice to Have

Engineering or industrial data background

Diagnostic / condition-monitoring experience

Mandatory Screening Questions

Please answer briefly:

Have you extended an existing analytics or ML system to support new anomaly types?

How would you add new anomaly detection without rewriting the system?

Confirm you understand this project is offline only, with no UI.