×

Industrial process fault detection using principal component analysis

  • US 6,952,657 B2
  • Filed: 09/10/2003
  • Issued: 10/04/2005
  • Est. Priority Date: 09/10/2003
  • Status: Expired due to Fees
First Claim
Patent Images

1. A method for monitoring an industrial process, said method comprising the steps of:

  • obtaining sensor data corresponding to a plurality of product units being processed in accordance with the industrial process;

    forming a sample matrix of data representing at least two of the product units, wherein the sample matrix is formed from at least a portion of the sensor data;

    computing a plurality of singular vectors of the sample matrix;

    reducing the plurality of singular vectors to a principal set of singular vectors;

    computing principal components of sensor data corresponding to at least one additional product unit processed subsequent to the product units represented in the sample matrix;

    computing a predicted data vector for the additional product unit;

    calculating a residual data vector for the additional product unit using the predicted data vector for the additional product unit and a measured data vector corresponding to the additional product unit, the measured data vector comprising sensor data obtained for the additional product unit;

    calculating a scalar metric from the residual data vector for the additional product unit; and

    categorizing the additional product unit based on the value of the scalar metric;

    wherein said steps of obtaining sensor data, forming a sample matrix, computing a plurality of singular vectors, reducing the plurality of singular vectors, computing principal components, computing a predicted data vector, calculating a residual data vector, calculating a scalar metric, and categorizing the additional product unit are performed in real time as additional product units are processed in accordance with the industrial process.

View all claims
  • 1 Assignment
Timeline View
Assignment View
    ×
    ×