Application of robust statistical models to measure data quality for improved use of sensors and diagnostics in an active mine setting

我们建议研究统计分析的新方法，以提高对工业传感器正常和异常行为的预测。能够在广泛的操作条件下检测细微变化，并衡量其对测量的潜在单个和复合影响，对于在具有挑战性的条件下部署传感设备的公司具有重大的潜在价值。在许多工业环境中，执行响应式传感器系统维护可能非常困难，因为它可能需要关闭或暂停操作。因此，仍然有机会处理受损的传感器数据是重要的，因为如果在常规操作程序中出现机会，**是预测何时应该进行预防性维护的机会。然而，自信地处理受损数据需要对整个系统有深入的了解，包括通常不被认为与传感模式相关的变量。MineSense Technologies Ltd.是一家位于不列颠哥伦比亚省的采矿技术公司，专门从事采矿各个阶段的多变量遥测技术。他们的客户是希望通过预先集中和分选开采矿石以及废弃岩石的排除来优化其**业务的矿业公司。用于开采矿石实时传感的算法开发是一项复杂的任务，需要与矿山工作人员密切讨论并共享数据和知识。以统计模式和评估为基础的机器学习方法的部署将是公司可以利用的另一种方法。利用统计学术专家的经验将加速(1)整合收集的各种类型的**数据和(2)探索先进统计研究可能带来的机会。这个项目的主要目标是：(1)描述和监测系统数据质量，(2)开发数据质量模型的科学和技术基础，以及(3)开发将这些模型集成到更广泛的系统健康监测中的方法。本研究结果有可能提供重要的参考

We propose to investigate novel approaches of statistical analysis to improve upon predictions of normal and**abnormal behavior in industrial sensors. The ability to detect subtle changes within a wide range of operational**conditions and gauge their potential individual and compounding influence on measurements has significant**potential value for companies deploying sensing devices in challenging conditions. It can be very difficult to**perform reactive sensor system maintenance in many industrial settings because it may require operational shut**down or suspension. Accordingly, the opportunity to still work with compromised sensor data is significant, as**is the opportunity to predict when preventative maintenance should be undertaken if an opportunity presents**itself within regular operational routines. Working confidently with compromised data, however, requires deep**understanding of an entire system, including variables perhaps not normally considered as relevant for the**sensing mode(s).**MineSense Technologies Ltd. is a B.C.-based mining technology company specializing in multivariate**telemetry of ore at all stages of mining. Their clients are mining companies that want to optimize their**operations through pre-concentration and sorting of mined ore as well as the rejection of waste rock. Algorithm**development for real-time sensing of mined ore is a complex task that requires intimate discussion and sharing**of data and knowledge with mine staff. Deployment of machine learning approaches with foundations in**statistical patterns and assessment would be another approach that the company could leverage. Harnessing the**experience of statistical academic experts would accelerate the process of (1) integrating the various types of**data being collected and (2) exploring opportunities that advanced statistical research could bring.**The main objectives of this project are to (1) characterize and monitor system data quality, (2) develop the**scientific and technical basis for data quality models, and (3) develop approaches to integrate these models**within broader system health monitoring. Results of the research have the potential to provide significant