# CLAUDE.md This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository. ## Project purpose Thermal analysis of boxes passing through a production oven (Cevolani Italia s.r.l.), measured with IR sensors. The pipeline accepts any `data.csv` that respects the column specifications in `README.md` and produces: - A Newton's cooling fit for each inductor zone (`induttoreX/appX.py`) - A professional PDF report (`report_completo.pdf`) aggregating all results The user directs the analysis (defines time windows, chooses exclusion zones); Claude implements the requested models and pipeline changes in Python. ## Setup ```bat python -m venv venv venv\Scripts\activate pip install -r requirements.txt ``` ## Project architecture ```text main.py ← interactive menu, auto-discovers induttore*/ folders start/ calcola_t_amb.py ← computes T∞ (weighted mean), writes start/report.md plot_temperatura.py ← plots raw temp_obj and temp_amb over [T0, T_END] induttoreX/ appX.py ← nonlinear fit for one inductor zone; params hardcoded by user report_completo.py ← reads all induttore*/report*.md, generates .tex + .pdf algoritmo.md ← detailed explanation of the TRF algorithm clear.sh ← deletes all generated files (*.md, *.png, report_completo.*) ``` ## Data `data.csv` required columns (exact names): - `time since start [ms]` — converted to seconds (`/1000`) in every script - `temp_obj IR [C]` — object/box temperature, the fitting target - `temp_amb IR [C]` — ambient temperature, used to compute T∞ T∞ is computed once by `calcola_t_amb.py` (weighted mean over the full acquisition) and stored in `start/report.md`. All other scripts read T∞ from there — it is **not hardcoded**. ## Fitting architecture All fits use `scipy.optimize.curve_fit(..., method="trf")` (Trust Region Reflective, nonlinear least squares). Newton's cooling model: ```text T(t) = T∞ + A · exp(-(t - t0) / τ) ``` Standard pattern in every `appX.py`: ```python T0, T_END = ... # time window [s], set by user W0_START, W0_END = ... # optional exclusion zone [s], set by user df = pd.read_csv(ROOT / "data.csv") df["time_s"] = df["time since start [ms]"] / 1000.0 mask = (df["time_s"] >= T0) & (df["time_s"] <= T_END) t_fit = df.loc[mask, "time_s"].values T_fit = df.loc[mask, "temp_obj IR [C]"].values sigma = np.where((t_fit >= W0_START) & (t_fit <= W0_END), 1e10, 1.0) mask_clean = sigma < 1e10 popt, pcov = curve_fit(model, t_fit, T_fit, p0=[...], bounds=([...], [...]), sigma=sigma, absolute_sigma=True, method="trf") residuals = T_fit[mask_clean] - model(t_fit[mask_clean], *popt) ss_res = np.sum(residuals**2) ss_tot = np.sum((T_fit[mask_clean] - np.mean(T_fit[mask_clean]))**2) R2 = 1 - ss_res / ss_tot perr = np.sqrt(np.diag(pcov)) ``` If no exclusion zone is needed, `sigma = np.ones(len(t_fit))` and `mask_clean = slice(None)`. ## Adding a new inductor 1. Copy an existing `induttoreX/appX.py` into a new folder `induttoreY/appY.py`. 2. Set `T0`, `T_END` to the desired time window. 3. Set `W0_START`, `W0_END` if a sub-interval must be excluded (or remove the exclusion logic). 4. Run via `main.py` — the menu discovers it automatically. 5. The script produces `induttoreY/induttoreY.png` and `induttoreY/reportY.md`. 6. Running `report_completo.py` includes it automatically in the PDF. ## Report generation `report_completo.py` is fully dynamic: - Discovers all `induttore*/report*.md` via glob - Reads T∞ from `start/report.md` - Renders equations and math symbols as PNG via matplotlib mathtext (no system LaTeX required) - Outputs `report_completo.tex` and `report_completo.pdf` Corporate identity constants (editable at the top of `report_completo.py`): `SOCIETA`, `DIPARTIMENTO`, `DOC_NUM`, `REVISIONE`, `STATO`, `CLASSIFICAZIONE`.