Baseline predictive maintenance method for target device and computer program product thereof

A baseline predictive maintenance method, comprising:

• monitoring a characteristic of a target device and a plurality of processing parameters related to the target device when the target device is processing a plurality of workpieces just after one maintenance has been performed on the target device, thereby obtaining a plurality of fresh samples, the target device including a throttle valve of a plasma-enhanced-chemical-vapor-deposition (PECVD) tool or a device of production equipment, the characteristic including an angle of the throttle valve or an operating parameter of the device, the processing parameters including pressure and RF power of the PECVD tool or related parameters of the production equipment, wherein the fresh samples are related to the target device and are generated when the target device produces the workpieces respectively, and the fresh samples are corresponding to the workpieces in a one-to-one manner;
• monitoring the characteristic of the target device and the processing parameters related to the target device when the target device is processing a new workpiece after the workpieces have been processed, thereby collecting a new workpiece sample, wherein the new workpiece sample is related to the target device and is generated when the target device produces the new workpiece, each of the fresh samples and the new workpiece sample comprising a set of paired data, the set of paired data consisting of a set of TD-related process data (X) and an actual value of the characteristic of the target device, the set of TD-related process data (X) comprising values of the processing parameters, wherein the target device is operated at one of the actual values of the characteristic of the target device and the values of the processing parameters are adopted by the production equipment when the target device is producing each of the workpieces and the new workpiece;
• building a TD baseline model by using a plurality of modeling samples in accordance with a conjecturing algorithm, wherein the modeling samples include the new workpiece sample and the fresh samples;
• and
• using the TD baseline model to compute a healthy baseline value (ŷB), wherein the healthy baseline value (ŷB) is a predicted value of the characteristic at which the target device should be operated when the target device is healthy and is producing the new workpiece;
• and
• based on the healthy baseline value (ŷB), performing fault-detection-and-classification on the target device, and performing another maintenance on the target device if the actual value of the characteristic of the target device in the new workpiece sample is deviated from the healthy baseline value of the target device by a threshold value when processing the new workpiece.