{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Plasma detection from $U_{loop}$\n", "\n", "This is a quick plasma start and end detection from the peaks in the derivative of the $U_{loop}$ signal.\n", "This procedure is run as soon as possible in order to provide the information on plasma existance to other routines." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "from scipy import signal # for peak detection\n", "import pandas as pd\n", "import requests\n", "import subprocess #workarround for db operation\n", "import os\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "shot_no = 37057" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def update_db_current_shot_alchemy(field_name, value):\n", " try:\n", " # local UNIX socket connection\n", " engine = sqlalchemy.create_engine('postgresql://golem@/golem_database?host=/var/run/postgresql')\n", " #engine = sqlalchemy.create_engine('postgresql://golem:rabijosille@192.168.2.116/golem_database')\n", " except:\n", " return\n", " engine.execute(f\"\"\"UPDATE shots SET \"{field_name}\"={value} WHERE shot_no IN(SELECT max(shot_no) FROM shots)\"\"\") \n", "\n", "def update_db_current_shot(field_name, value):\n", " os.system('export PGPASSWORD=\"rabijosille\";psql -c \"UPDATE shots SET '+field_name+'='+str(value)+'WHERE shot_no IN(SELECT max(shot_no) FROM shots)\" -q -U golem golem_database')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "os.makedirs('Results', exist_ok=True)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def save_scalar(phys_quant, value, format_str='%.3f'):\n", " with open(\"Results/\"+phys_quant, 'w') as f:\n", " f.write(format_str % value)\n", " update_db_current_shot(phys_quant,value) \n", " " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "t_CD = requests.get(\n", " f'http://golem.fjfi.cvut.cz/shots/{shot_no}/Production/Parameters/Tcd')\n", "t_CD = float(t_CD.content) * 1e-3 # from us to ms\n", "t_CD" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "U_loop = pd.read_csv(f'http://golem.fjfi.cvut.cz/shots/{shot_no}/Diagnostics/BasicDiagnostics/U_Loop.csv',\n", " names=['Time', 'U_loop'], index_col='Time', squeeze=True)\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The processing is done only beyond the CD system activation time (to prevent thyristor noise from e.g. the $B_t$ system)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "U_loop.index *= 1e3 # s to ms " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "U_loop = U_loop.loc[t_CD:]" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "dt = np.diff(U_loop.index[[0,-1]]).item() / U_loop.index.size\n", "dt" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def to_points(time, must_be_odd=False):\n", " n = int(np.rint(time / dt))\n", " if n % 2 == 0 and must_be_odd:\n", " n += 1\n", " return n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "Estimate $\\frac{d U_{loop}}{dt}$ by a second order Savitzky-Golay 1. derivative filter." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "Dt = 0.1 # window length in ms\n", "dU = pd.Series(signal.savgol_filter(U_loop.loc[t_CD:], to_points(Dt, must_be_odd=True),\n", " 2, 1, delta=dt), index=U_loop.index, name='dU_dt')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "def argrelmax_time(sig, Dt, res_index, threshold_value=None, \n", " comparator=np.greater):\n", " idxs = signal.argrelextrema(sig.values, comparator, order=to_points(Dt))\n", " sig_sel = sig.iloc[idxs]\n", " if threshold_value is not None:\n", " sig_sel = sig_sel[comparator(sig_sel, threshold_value)]\n", " return sig_sel.index[res_index]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The plasma starts when the $U_{loop}$ decreases significantly for the first (index 0) time." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "try:\n", " t_plasma_start = argrelmax_time(dU, Dt/2, 0, threshold_value=-3,\n", " comparator=np.less)\n", "except IndexError:\n", " t_plasma_start = -1.0 # not found, nan convention\n", "t_plasma_start" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "The plasma ends when the $U_{loop}$ increases significantly for the last (index -1) time." ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "try:\n", " t_plasma_end = argrelmax_time(dU.loc[t_plasma_start:], Dt/2, -1,\n", " threshold_value=10,\n", " comparator=np.greater)\n", "except IndexError:\n", " t_plasma_end = -1.0 # not found, nan convention\n", "t_plasma_end" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "is_plasma = int(t_plasma_start > 0 and t_plasma_end > 0)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "if is_plasma:\n", " t_plasma_duration = t_plasma_end - t_plasma_start\n", "else:\n", " t_plasma_duration = -1.0 # convention instead of nan" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "print(f'plasma: {bool(is_plasma)}, from {t_plasma_start:.1f} to {t_plasma_end:.1f} ms (lifetime {t_plasma_duration:.1f} ms)')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "save_scalar(\"is_plasma\", is_plasma)\n", "save_scalar(\"t_plasma_start\", t_plasma_start)\n", "save_scalar(\"t_plasma_end\", t_plasma_end)\n", "save_scalar(\"t_plasma_duration\", t_plasma_duration)" ] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.7.4" } }, "nbformat": 4, "nbformat_minor": 4 }