基于精简并发潜结构映射的竖炉焙烧过程综合故障诊断

竖炉焙烧过程因运行条件异常变化或操作不当会造成上火、冒火、过还原和欠还原等运行故障. 这些故障直接影响过程运行安全和产品质量 (比如, 磁选管回收率), 但难以采用基于模型和基于知识的方法建模故障与产品质量的关系, 以及诊 断故障变量. 针对上述问题, 本文提出数据驱动的基于并发潜结构映射 (Concurrent projection to latent structures, CPLS) 的竖炉焙烧过程综合故障诊断方法. 首先, 将并发潜结构映射分解的过程变量共有子空间与残差空间精简合并来建立磁选管回收率相关的过程变化空间, 提出基于精简并发潜结构映射模型的竖炉焙烧过程综合监控方法; 接下来, 定义相应的重构贡献图并与竖炉焙烧过程相结合, 提出 CPLS 精简重构贡献方法用于竖炉焙烧过程故障变量诊断; 最后, 利用竖炉焙烧过程半实物仿真平台采集的数据进行实验研究, 结果表明所提方法不仅可以诊断出质量相关的故障, 而且可诊断出回路设定值之外的故障变量.

豎爐焙燒過程因運行條件異常變化或操作不當會造成上火、冒火、過還原和欠還原等運行故障. 這些故障直接影響過程運行安全和產品品質 (比如, 磁選管回收率), 但難以採用基於模型和基於知識的方法建模故障與產品品質的關係, 以及診 斷故障變數. 針對上述問題, 本文提出資料驅動的基於併發潛結構映射 (Concurrent projection to latent structures, CPLS) 的豎爐焙燒過程綜合故障診斷方法. 首先, 將併發潛結構映射分解的過程變數共有子空間與殘差空間精簡合併來建立磁選管回收率相關的過程變化空間, 提出基於精簡併發潛結構映射模型的豎爐焙燒過程綜合監控方法; 接下來, 定義相應的重構貢獻圖並與豎爐焙燒過程相結合, 提出 CPLS 精簡重構貢獻方法用於豎爐焙燒過程故障變數診斷; 最後, 利用豎爐焙燒過程半實物模擬平臺採集的資料進行實驗研究, 結果表明所提方法不僅可以診斷出品質相關的故障, 而且可診斷出回路設定值之外的故障變數.

Operational faults of shaft furnace roasting processes can appear when operational conditions change abnormally or operators do not react properly or timely. Typical operational faults, including fire-emitting, flame-out, under-reduction and over-reduction, are highly related to process safety and product quality, e.g., magnetic tube recovery rate (MTRR). Fault diagnosis of shaft furnace roasting processes deserves more attentions. However, it is difficult to apply model-based or knowledge-based fault diagnosis methods. In particular, it is difficult to model the relations between fault and product quality. In this paper data-driven concurrent projection to latent structures (CPLS) based fault diagnosis is developed for shaft furnace roasting processes. First, a CPLS based comprehensive monitoring method for shaft furnace roasting processes is proposed by combining co-variation and residual of process spaces of concurrent projection to latent structures into a simplified MTRR-relevant process-variation space. Secondly, a corresponding simplified reconstruction-based contribution method is proposed and used to pinpoint the faulty variable. Finally, the proposed methods are verified using the data collected from a hardware-in-loop simulation platform. The results demonstrate that the quality-relevant faults as well as faulty variables are successfully diagnosed.