通过使用适合加速度信号的多元自回归模型，而不是单变量模型，可以获得接近故障的早期指示。从端铣寿命试验数据可以看出，多变量模型可以比单变量模型更早地识别即将发生故障的迹象；对于所呈现的情况，在失效前30英寸，而不是单变量模型获得的6.5英寸。这一额外的警告时间允许采取预防措施，并允许完成当前切割，以便在切割之间更换刀具。

By using multivariate autoregressive models fit to acceleration signals, instead of univariate models, an earlier indication of approaching failure is obtained. From end-milling life tests data, it is demonstrated that the multivariate models can identify the indications of impending failure earlier than is possible using univariate models; for the case presented, 30 inches prior to failure, as opposed to the 6.5 inches obtained with the univariate models. This extra warning time, allows for preventive action to be taken and allows the possibility of finishing the current cut so that the tool change can be made between cuts.