Time series stationarize vs normalization

I have multiple time series coming from sensor measurements of an industrial machine. The industrial machine runs different ‘Recipes’. Every recipe has different set of parameters which are set before a recipe begins. So this means across time my sensor measurements will be affected when a new recipe begins. So lets say for the first 1 hour I run my recipe which gives me temperature measurments of 80 degree celcius, then i change to another recipe and run it for 2nd hour which gives me temperature as 90 degree celcius.

In total I have a lot of recipes running. So eventually my timeseries has varying mean and variance across time which comes from running different recipes. My end goal is to predict when the machine will breakdown. So eventually the features go into some machine learning model.

Before feeding my sensor measurements to the machine learning model i must do some sort of normalization or maybe also stationarize my timeseries.

1. Do you recommend normalizing every part of my timeseries ( by every part i mean every recipe window of my timeseries ) or stationarize the timeseries ?

For normalization I can simply put in a standard scaler for every recipe but to stationarize a timeseries i find it bit difficult especially because I dont see any kind of seasonality or trend in my timeseries. But still the timeseries is non stationary ( because mean and variance varies over time due to changes in recipes ).

2. So how should I stationarize my timeseries ?