PyLCSS: Low-Code System Solutions

Overview

https://github.com/kutaydemir07/PyLCSS PyLCSS (Python Low-Code System Solutions) is a professional engineering design platform. It allows engineers to model complex multidisciplinary systems through a node-based visual interface, run parametric CAD and FEA simulations, explore high-dimensional Solution Spaces, and optimise designs using 7 different algorithms — all within a single desktop application.

Built for real-world engineering workflows, PyLCSS features a crash-free multi-threaded architecture, vectorised computation kernels, comprehensive file I/O, and an integrated AI coding assistant.

pylcss

Scientific Foundation

PyLCSS implements the Solution Space approach for robust design: instead of seeking a single optimal point (which may be sensitive to tolerances), it identifies box-shaped regions of valid designs, enabling decoupled subsystem development.

Markus Zimmermann, Johannes Edler von Hoessle, “Computing solution spaces for robust design”, Int. J. Numer. Meth. Engng., 2013. DOI: 10.1002/nme.4450

Key Features

Parametric CAD Environment

70+ Node Types — Primitives, Booleans, Fillets, Chamfers, Sweeps, Lofts, Shells, Patterns, Imports
Topology Optimisation — SIMP with MMA/OC solvers, density/sensitivity filtering, Heaviside projection, symmetry constraints, shape recovery with marching cubes, and direct STL/OBJ export of optimised shapes
Advanced Nodes — Thicken, Pipe, Split, Text emboss, Math Expression evaluator, Import STEP/STL
Real-Time 3D Viewer — VTK-based with density cutoff preview during optimisation
Measurement — Distance, surface area, and volume nodes

Finite Element Analysis (FEA)

scikit-fem + Netgen meshing — Tetrahedral/triangular elements
Linear Elasticity — Displacement, von Mises stress, compliance
FEA Results Nodes — Stress extraction, displacement, reaction forces
Remeshing — Surface-to-solid conversion for topology-optimised shapes (up to 20 000 faces)

Multi-Objective Optimisation (7 Solvers)

Algorithm	Type	Best For
SLSQP	Gradient-based	Fast local optimisation with constraints
COBYLA	Derivative-free	Noisy or non-differentiable models
trust-constr	Interior point	Large-scale constrained problems
Differential Evolution	Population-based	Global search, black-box functions
Nevergrad	Meta-optimiser	Algorithm-agnostic global search
NSGA-II	Multi-objective evolutionary	Pareto fronts with 2–5 objectives
Multi-Start	Hybrid global+local	Avoiding local minima via LHS starts

Global Sensitivity Analysis (4 Methods)

Method	Indices	Use Case
Sobol	S1, ST, S2 interaction	Variance decomposition
Morris	μ, μ*, σ	Screening with few evaluations
FAST	S1, ST	Fourier decomposition, fast convergence
Delta (DMIM)	δ, S1	Moment-independent, distribution-based

Batch analysis across all outputs, convergence study, importance ranking (Critical / Important / Minor / Negligible)

Surrogate Modelling & Validation

5 Algorithms — MLP Neural Network (PyTorch), Random Forest, Gradient Boosting, Gaussian Process, SVR
Cross-Validation — K-Fold (2–20 folds) and Leave-One-Out
Model Comparison — Automated comparison of all 5 algorithms on same dataset
Feature Importance — Permutation-based and tree-based importance analysis
Hyperparameter Optimisation — Grid search and random search with built-in search spaces

Solution Space Exploration

Monte Carlo Sampling — Vectorised evaluation of thousands of design variants
Visualisation — 2D scatter, parallel coordinates, feasibility maps
Product Family Analysis — Common platform identification across product variants
Step Analysis — Iterative box-size refinement

Export Capabilities

Category	Formats
CAD Export	STEP, STL, OBJ
Data Export	CSV, Images (PNG, SVG, PDF)
Project	Project Folder (JSON-based `.cad` files + data)

Engineering Utilities

Expression-Aware Inputs — Safe AST-based evaluation (sin, cos, sqrt, log, variables) in input fields
Unit-Aware Inputs — Physical unit support via pint (SI, Imperial, CGS) in simulation nodes

LLM-Powered Voice Assistant

Natural Language Control — “Zoom in”, “Create a helical gear”, “Go to properties”
Local STT — Faster-Whisper for real-time speech recognition
Multi-Provider LLM — OpenAI, Claude, Gemini, LM-Studio
Privacy-First — Optional fully local execution

Installation

Prerequisites

Python 3.8+
OS Windows 10/11 (macOS and Linux: experimental)

Quick Install

# Clone
git clone <repository-url>
cd pylcss

# Virtual environment
python -m venv .venv
# Windows:
.venv\Scripts\activate
# Linux/Mac:
source .venv/bin/activate

# Dependencies
pip install -r requirements.txt

# Launch
python scripts/main.py

Or on Windows: double-click run_gui.bat.

Quick Start

Launch — python scripts/main.py
Load a Model — File → Open → select data/Gear Unit.json
Validate — Click “Validate” to check units and connections
Solution Space — Switch to Solution Space tab → “Compute”
Visualise — Plot Weight vs. Safety Factor
Optimise — Go to Optimisation tab → select objectives → Run

Architecture

pylcss/
├── cad/                  # Parametric CAD kernel (CadQuery + OCC)
│   ├── nodes/            # 50+ node types (primitives, booleans, FEA, TopOpt)
│   ├── engine.py         # Graph execution engine
│   └── node_library.py   # Node registry
├── optimization/         # 7 solvers (SciPy, Nevergrad, NSGA-II, Multi-Start)
├── sensitivity/          # 4 methods (Sobol, Morris, FAST, Delta)
├── solution_space/       # Monte Carlo, step analysis, product families
├── surrogate_modeling/   # 5 ML algorithms + CV + HPO + feature importance
├── io_manager/           # CAD/mesh/data/project I/O (15+ formats)
├── system_modeling/      # Graph-based system model builder
├── assistant_systems/    # LLM voice assistant & tools
└── user_interface/       # PySide6 + VTK desktop application

Tech Stack

Layer	Technologies
UI	PySide6, NodeGraphQt, QtAwesome
CAD	CadQuery, OpenCASCADE (OCP), VTK
FEA	scikit-fem, Netgen, meshio
Computation	NumPy, SciPy, Pandas
Visualisation	VTK (3D), pyqtgraph (2D)
ML	PyTorch, scikit-learn
Optimisation	SciPy, Nevergrad, SALib
Units	pint
Serialisation	h5py, dill, joblib
AI Assistant	Faster-Whisper, OpenAI, Edge-TTS

License

Licensed under the PolyForm Shield License 1.0.0.

Allowed: Personal use, academic research, internal business use. Restricted: You cannot use this software to build a competing product or service.

See LICENSE for full details.