all very good theories....which is correct? It might be obvious, but then why would the OP ask?
It might be as simple as determining what's wrong: is anything out of spec or obviously misprocessed?
We could ask what changed but this is often difficult to assess before we understand the causal mechanism as everything changes...and we rarely measure andrecord all of the input factors (material, process settings and conditions) to correlate to breaks. A better start is to ask what patterns exist? What is the rate? Is it all of the pieces or only some? Is there a time element to this breaking? Did it suddenly start, or has it 'always' been there at the current rate? Does it come and go, is what is the pattern: plot the data in time series it will tell you quite a bit.
We could ask what's different: is there any feature, dimension or property that is at one level for the parts that break and another level for the parts that don't break? Are there different material lots involved? Different operators? Different machines?
We could go fully functional: you are exceeding the strength of the material as Golfman says. But it it the material or the machine? Can you measure the strength of the material and the machine separately from each other?