Statistical Pattern Recognition for Manufacturing Quality Control

统计过程控制(SPC)在控制产品质量中起着重要作用。当过程平均值或产品变异性偏离允许范围时，SPC发出信号**。这种区域测试，也称为运行测试，是被广泛接受的识别异常过程行为的经典技术。**然而，在规范限制内运行并不一定意味着过程稳定。运行测试已被证明**在指示失控情况方面是有效的，但其在解释过程数据方面的有效性**仍存在争议。**本提案的主要目标是对此类定制检验应用程序进行培训，以识别**完美和有缺陷的项目的样本。这也是Sightline公司提高其**市场竞争力的战略目标。一旦计算模型达到足够的精确度，它将被认为**准备好开始检查并部署到生产环境中。在接收到新数据后，**模型将结合机器学习算法使用统计模式识别来检测**产品中的任何差异，如划痕、污染物或不正确的尺寸。**如果产品合格或检测到缺陷，将立即通知操作员。随着时间的推移，该模型接收到**高质量的检验数据，可用于预测性分析，从而**了解缺陷何时、何地、如何和为什么进入工艺，从而提供充分的机会**采取先发制人和积极主动的措施。这个项目的最终目标是开发机器**学习模型，该模型可以自动学习从制造过程中收集的大型图像和传感器**数据集中的复杂结构和趋势。使用这些习得的特征，我们可以识别新制造产品中的已知缺陷**，并有可能发现新的缺陷。

Statistical process control (SPC) plays an important role in controlling a product's quality. SPC gives a signal**when the process mean or the product variability moves away from the permissible range. Such zone tests, also**known as run tests, are widely accepted classical techniques for recognition of the abnormal process behaviour.**However, operating within the specification limits does not necessarily signify a stable process. Run tests have**proven effective in indicating out-of-control situations, their effectiveness in interpreting process data,**however, is disputed.**The prime goal of this proposal is to have such custom inspection applications trained to recognize samples of**perfect and defective items. This is also the strategic objective of the Sightline company for increasing its**market competitiveness. Once the computational model reaches a sufficient level of accuracy, it would be**deemed ready to start inspecting and be deployed in a production environment. Upon receiving new data, the**model would use statistical pattern recognition alongside machine learning algorithms to detect any variance in**the product, such as scratches, contaminants, or improper dimensions.**If the product passes or a defect is detected, the operator would be notified immediately. As the model receives**high quality inspection data over time, it can be used for predictive analysis which would allow one to**understand when, where, how, and why defects are entering the process, thereby providing ample opportunity**to take preemptive and proactive measures. The ultimate objective of this project is to develop machine**learning models which can automatically learn complex structures and trends in large image and sensor**datasets collected from manufacturing processes. Using these learned features, we can identify known defects**in newly manufactured products and, potentially, discover new ones.