Abstract Low-analysis fertilizer composition highly poses risk to soil infertility, crop yield and farmer livelihood. Fertilizer testing via traditional laboratory methods is a tool for quality control but with the significantly high cost of investment and maintenance in the long run. Physiochemical data from vibrational spectra obtained from the near-infrared reflectance (NIR) analyzer shows successful installation design for real-time quality monitoring of compound fertilizer production in this study overcoming exhaustive laboratory methods offering competitive advantages to the fertilizer industry becoming a great tool for product development and lower the risk of farmers buying the low-analysis fertilizer. The NIR analyzer was designed for on-line analysis of nutrient contents of the fertilizer before bagging. Total nitrogen (TN%), available phosphorus (available P2O5%), water-soluble potassium (water-soluble K2O%), total organic matter (TOM%), moisture content (TMC%) including pH were required to be monitored to ensure final product release being within limits legislatively established. Spectral variations across different times of measurement over a day of the production was investigated in the condition of on-line system instead of a controlled condition to confirm the measurement will be reliable and in good reproducibility. Calibration models were developed using Partial Least Squares (PLS) regression with a series of data pretreatments on individual model per nutrient. Precision and accuracy figures of NIR results were obtained considerably good enough to the manufacturing practice. NIR prediction results were later designed for threshold and feed-back to the process control.

中文翻译：

---

复合肥生产在线近红外光谱仪安装设计

摘要 低分析肥料成分对土壤贫瘠、作物产量和农民生计构成高度风险。通过传统实验室方法进行的肥料测试是一种质量控制工具，但从长远来看，投资和维护成本非常高。从近红外反射 (NIR) 分析仪获得的振动光谱的理化数据表明，本研究中复合肥生产实时质量监测的安装设计成功克服了详尽的实验室方法，为肥料行业提供了竞争优势，成为产品的重要工具发展并降低农民购买低分析肥料的风险。NIR 分析仪设计用于在装袋前在线分析肥料的养分含量。总氮 (TN%), 需要监测有效磷（有效 P2O5%）、水溶性钾（水溶性 K2O%）、总有机物（TOM%）、水分含量（TMC%），包括 pH 值，以确保最终产品的释放量在立法限制范围内已确立的。在在线系统的条件下而不是在受控条件下研究一天生产中不同测量时间的光谱变化，以确认测量将可靠且具有良好的重现性。校准模型是使用偏最小二乘法 (PLS) 回归开发的，其中对每种营养素的单个模型进行了一系列数据预处理。获得的 NIR 结果的精度和准确度数据对于制造实践来说已经足够好了。