面向炭化产线的秸秆原料成分检测模块设计

Design of Straw Raw Material Component Detection Module for Carbonization Production Line

摘要:

秸秆作为我国最主要的农林废弃物，是制备生物炭的主要原料，秸秆原料的固定碳、挥发分、灰分含量是影响成炭品质的关键指标，也是炭化热解工艺参数的调控依据。针对当前秸秆炭化产线上对秸秆固定碳、挥发分、灰分成分含量的快速检测需求，设计了一种秸秆原料成分检测模块。首先基于近红外光谱技术开展秸秆炭化成分检测方法的研究，基于在线检测需求选择便携式光谱传感器并设计漫反射检测光路，搭建面向产线的光谱采集单元，采集粗切秸秆在1100~2500 nm范围内的漫反射光谱，结合Savitzky-Golay卷积平滑 （SG）、多元散射校正（MSC）、标准正态变换（SNV）预处理方法和偏最小二乘法（PLS），分别建立了基于全波长和基于竞争性自适应重加权法（CARS）筛选的特征波长的秸秆固定碳、挥发分、灰分含量预测模型，结果表明特征波长的建模效果优于全波长，对固定碳、挥发分、灰分质量分数预测的最优模型分别为SG+MSC-CARS-PLS、SG-CARS-PLS、SG+MSC-CARS-PLS，测试集决定系数R2分别为0.891 6、0.931 7、0.929 7，预测集均方根误差分别为1.46%、1.39%、0.42%，相对分析误差分别为2.54、3.44、3.18，能够实现精确预测。然后进行秸秆成分在线检测模块设计，模块分为光谱采集单元、供电单元、控制与传输单元，嵌入已构建的秸秆成分预测模型，基于Raspberry Pi 4B开发板和其自带的Wi-Fi模块实现秸秆在线光谱采集、模型计算、数据传输等功能，通过样机试验证明该模块设计及开窗位置选择可以采集满足在线分析要求的近红外光谱曲线，同时采用斜率/截距校正的方法，将实验室模型转移到产线进行在线应用，固定碳、挥发分、灰分含量预测精度均得到提升，可以达到在线分析需求，为热解工艺参数的调控提供数据支撑。

Abstract:

Straw, as the primary agricultural waste in China, is the main material for biochar production. The fixed carbon, volatile matter, and ash content are key indicators that influence biochar quality and guide the parameters of pyrolysis processes. Aiming to address the need for rapid detection of these indicators in straw carbonization production lines by designing a straw composition detection module. Utilizing near-infrared spectroscopy, a method for detecting straw components was developed. A portable spectral sensor was selected and a diffuse reflectance detection path was designed. A spectral collection unit was established to capture spectra of coarse-cut straw in the 1100~2500 nm range. By applying Savitzky-Golay convolution smoothing（SG）, multiple scattering correction （MSC）, standard normal variate （SNV） preprocessing, and partial least squares regression （PLS）, quantitative prediction models were developed for fixed carbon, volatile matter，and ash content by using full wavelengths and feature wavelengths selected by competitive adaptive reweighted sampling （CARS）. Results indicated that models based on feature wavelengths outperformed those using full wavelengths. The optimal models for fixed carbon, volatile matter, and ash content were SG+MSC-CARS-PLS, SG-CARS-PLS, and SG+MSC-CARS-PLS, with prediction set correlation coefficients R2 of 0.891 6，0.931 7, and 0.929 7, respectively. The root mean square errors of prediction set （RMSEp） were 1.46%，1.39%, and 0.42%, yielding relative prediction deviations （RPD） of 2.54, 3.44, and 3.18, demonstrating accurate prediction capabilities. Furthermore, an online detection module for straw composition was designed. The module was divided into three units： the spectroscopic acquisition unit, the power supply unit, and the control and transmission unit. The pre-built model for predicting straw composition was embedded in the module. Based on the Raspberry Pi 4B development board and its built-in Wi-Fi module, it enabled functions such as online spectroscopic acquisition, model computation, and data transmission for straw. Through prototype testing, it was demonstrated that the module design and window location selection could capture near-infrared spectroscopic curves that met the requirements for online analysis. By using the slope/intercept calibration method, the laboratory model was transferred to the production line for online application. The prediction accuracy of fixed carbon，volatile matter, and ash content was improved, fulfilling the requirements for online analysis and providing data support for the regulation of pyrolysis process parameters.