📄 中文摘要
异常和故障检测方法在识别系统操作条件的偏差中至关重要，这使得能够提前采取措施，通常可以防止更严重的损害。持续的偏差通常表明故障，而数据中的突发、孤立变化则指示异常。然而，在许多实际应用中，数据的变化并不总是代表异常的系统状态。这些变化可能被错误地识别为故障，而实际上是系统正常演变的表现，例如，处理新产品的开始特征，即实现领域转变。因此，区分故障与这种“健康”数据分布变化之间的差异，对于确保系统的稳定性和可靠性至关重要。
区分工业数据流中的故障与领域转变
📄 English Summary
Towards Differentiating Between Failures and Domain Shifts in Industrial Data Streams
Anomaly and failure detection methods are essential for identifying deviations from normal operational conditions, enabling proactive measures to prevent more severe damages. Long-lasting deviations typically indicate failures, while sudden, isolated changes in data signify anomalies. However, in many practical applications, changes in data do not always reflect abnormal system states. Such changes may be incorrectly identified as failures when they are actually a normal evolution of the system, such as the characteristics associated with starting the processing of a new product, referred to as a domain shift. Therefore, distinguishing between failures and these 'healthy' changes in data distribution is critical for ensuring system stability and reliability.