Multi-Vari Analysis: Locating the True Source of Process Variation

A production manager called me at six in the morning. His voice was tight. For three weeks, his plant had been fighting diameter variation on a critical cylindrical component. The tooling engineers blamed the material. The material engineers blamed the tooling. The operators just shrugged and said they were following the standard work.

I have heard this exact story in aerospace assembly halls, automotive Tier 1 plants, and foundries. A variability problem surfaces, and every department points a finger at the next. The typical response is a flurry of unstructured trial-and-error adjustments that temporarily mask the symptom before it returns two weeks later.

I told the manager to send me the data and prepare access to the line. When a process is unstable, an X-bar R chart confirms that you have a problem. It does absolutely nothing to tell you where the problem lives. For that, you need to break the total variation down into its fundamental components using Multi-Vari analysis.

The Three Dimensions of Variation

Multi-Vari analysis is a diagnostic method developed by Leonard Seder in the 1950s. Its objective is to decompose total process variation into three distinct dimensions to determine which one contributes most heavily to the observed scatter. It replaces departmental arguments with visual evidence.

The first dimension is within-piece variability. This captures differences across a single part or component. The second is piece-to-piece variability, which measures differences between consecutive parts produced under the same nominal conditions. The third is time-to-time variability, which tracks shifts between machine setups, operator changes, shifts, or days.

The power of this method is not mathematical complexity. It is visual clarity. When you construct a Multi-Vari chart correctly, the answer to where your problem originates jumps off the page. Everyone in the room, from the operator to the plant manager, reaches the same conclusion simultaneously.

Designing the Sampling Plan

When I arrived at the plant, the quality engineer showed me the X-bar R chart with pride. I asked her to read it again. Instability tells you that something is wrong; it does not tell you where. To find the root cause, we needed a structured sampling plan rather than ad-hoc data pulls from the quality database.

I designed a simple study. We would take measurements across three consecutive days. For each day, we would sample at three time intervals: start of shift, mid-shift, and end of shift. This structure captured the time-to-time dimension comprehensively without disrupting production.

Within each time interval, we pulled five consecutive parts. On each part, we measured the internal diameter at three specific positions: the front, the middle, and the back of the cylinder. Five parts multiplied by three positions yields fifteen data points per interval. Across three intervals and three days, we had 135 total measurements. No expensive software or complex Design of Experiments was required.

Constructing the Multi-Vari Study

1. 01Define the characteristicSelect one critical dimension, weight, or parameter to track.
2. 02Establish time intervalsSample across different shifts, days, or machine startups.
3. 03Sample consecutive piecesPull 3 to 5 sequential parts within each defined time interval.
4. 04Measure multiple positionsTake 3 distinct readings across each individual part.
5. 05Plot the chartMap vertical lines for within-piece variation, horizontal for piece-to-piece.

Data Collection on the Shop Floor

A Multi-Vari chart is only as reliable as the data it is built upon. I went to the line to watch the operator take the measurements. I immediately noticed three issues that would have invalidated the entire study if left uncorrected. The data collection method was actively obscuring the process reality.

First, the bore gauge had not been calibrated in three months. Second, the operator was only measuring the middle of the cylinder, ignoring the front and back positions entirely. Third, the results were being recorded in a notebook with a pencil, allowing for subjective rounding and erasing. The current data was effectively useless for diagnosis.

We calibrated the gauge immediately. I printed a structured data collection sheet with the three measurement positions clearly marked on a schematic of the cylinder. We mandated that all entries be made in pen. Before we even plotted the chart, tightening the measurement system alone improved our visibility into the process.

Reading the Visual Evidence

With clean data in hand, we plotted the chart. On the X-axis, we laid out the time periods: the days and the shift intervals. At each interval, we grouped the five consecutive parts. Each part was represented by three dots connected by a vertical line, illustrating the within-piece variation across the cylinder's length.

The within-piece variation was minimal. The three points on each cylinder clustered tightly together. The tooling was machining the cylinder consistently along its entire length. Furthermore, the piece-to-piece variation among the five consecutive parts was also relatively small. The parts coming out of the machine in a given run were practically identical.

The time-to-time variation was massive. There was a sharp jump in the mean diameter between the start of the shift and the mid-shift sample. The problem was not the material, the tooling, or the operator's competence. The problem was the machine setup process.

Multi-Vari does not tell you why the process failed. It tells you exactly where to point the microscope.

Every machine startup required a new adjustment, and those adjustments were entirely uncontrolled. One operator set the hydraulic pressure at 180 bar, another at 195. Sometimes the tooling was warmed up to operating temperature; often it was started cold. The Multi-Vari chart isolated the instability to the startup routine, turning a general complaint into a specific engineering target.

Targeting the Root Cause

With the variability source identified, we logged the actual setup parameters at each startup. We quickly identified a direct correlation between the tooling temperature at the moment of startup and the resulting cylinder diameter. A cold machine produced parts significantly out of tolerance.

We implemented a standard warm-up cycle. The tooling had to sit at operational temperature for a minimum of twenty minutes before the first part could be run. We paired this with a visual setup checklist requiring the operator to verify and log pressure, temperature, and cycle speed before every shift.

Impact of Standardized Setup

0.87Initial CpkHighly unstable and incapable of meeting automotive requirements.

1.52Final CpkAchieved within two weeks by regulating tooling temperature.

20 minWarm-up cycleThe critical setup standard missing from the original process.

The internal diameter scatter dropped from a Cpk of 0.87 to a Cpk of 1.52 within two weeks. We achieved this without capital investment, without changing the material supplier, and without regrinding the tooling. We simply systematized the setup process because the data finally told us where the instability lived.

When the Method Falls Short

Multi-Vari is a diagnostic first step, not a complete problem-solving toolkit. It highlights where to look, but it will not give you the chemical or mechanical reason for the failure. Once the chart identifies the dimension of variation, you must immediately deploy standard root cause tools like 5 Why, Ishikawa diagrams, or a formal Design of Experiments.

Occasionally, a Multi-Vari chart will show heavy scatter across all three dimensions simultaneously. When within-piece, piece-to-piece, and time-to-time variations are equally bad, you almost certainly have a measurement system failure. Before you waste engineering hours analyzing the process, verify your Measurement System Analysis (MSA). If your Gage R&R is above 30 percent, the chart is only visualizing gauge noise.

This is why I insist on validating the MSA before launching any Multi-Vari study. Building an analysis on unverified data is constructing a house of cards. Trustworthy measurements are the absolute prerequisite for visual process diagnosis.

Cross-Functional Communication

The most underappreciated strength of Multi-Vari analysis is its ability to align a room. When I drew that chart on a flipchart in front of fifteen operators, engineers, and managers, I did not need a thirty-minute PowerPoint briefing. One look at the clusters, and everyone understood that the setup was the issue.

The best quality engineering methods are not necessarily the most mathematically sophisticated. They are the ones that create a shared language between people who usually speak different professional dialects. A Multi-Vari chart answers the question everyone has been secretly asking for weeks: is this actually my department's fault?

When the chart clears a department, the defensiveness drops. People stop pointing fingers at the neighboring workstation. The team focuses its energy on the single point of failure highlighted by the data, and the problem-solving accelerates immediately.