Breakthrough Improvements in Biscuit Manufacturing

           

Abbreviations
t	Thickness of stack of product
T	Batch average of t sampled in the hour
Ta	Average if batch averages T
Tp	Population average of T
Tst	Desired value of T (and Tp)
S100	Index of sigma of batch averages over 100 hours, treated as sigma of population of
x1 … x11	11 input variables determining stack thickness
I	Intercept in multiple regression
a1..a11	Coefficient of x1… x11 in multiple regression
X11	Batch average of x11 sampled in the hour
Ss100	Index of sigma of batch averages of x11 over 100 hours, treated as sigma of population X11*

Recognizing that quality is a key growth driver in an increasingly competitive market, the new management of a large, well-established FMCG food company in India decided to explore total quality management (TQM) as the means of establishing a quality culture. The case presented here is the first successful application of TQM in the company’s manufacturing processes. The dramatic improvements achieved went a long way in demonstrating that TQM works.

The problem
Damaged stock returned from the market (0.76 percent of sales) was a chronic problem defying resolution, and senior management threw this challenge to TQM. A quick analysis indicated that 67 percent of the returns were due to broken biscuits.

Reducing breakage in a specific product line was therefore chosen as the first project. A cross-functional factory team was selected for the effort. The manufacturing process has the following key steps:

1. Mix ingredients as per the recipe
2. Preprocess into a pliable mass
3. Sheet mass to the specified thickness
4. Mold-cut the sheet into individual units
5. Cook in a six-stage oven
6. Pack stacks of products into flexible retail packs
7. Packing into shippers for dispatch

A two-day quality training program was conducted with the team to introduce them to TQM, why and how it works, and most important, to open their minds to explore change. The project was then begun using TQM’s seven-step problem-solving method, as defined below:

Step 1	Define the problem	Problem = desire – current status
Step 2	Root cause analysis	Why? Why? Why? Why? Why?
Step 3	Generate countermeasure ideas
Step 4	Test the ideas and implement them in production
Step 5	Check the result
Step 6	Standardize procedures
Step 7	Prepare the quality-initiative story

Step 1—Define the problem
Using the TQM definition "Problem = desire – current status" data revealed that the breakage level was 0.53 percent. Given the intractable nature of the problem, management said they would be delighted with a reduction of 50 percent in the factory breakage levels.

Step 2 —The root cause analysis
Observation of the line and a quick scan of the data revealed maximum breakage occurred while feeding a stack of products into the packing machine conveyor. Experience attributed this to varying stack lengths. Simultaneous hourly sampling of stack thickness (t) and percentage breakage for three-minute periods represented in a histogram of the variation of t (represented by its sigma) and breakage confirmed this relationship (See Figure 1).

Figure 1

(Note: sigma values are used as comparative measures of improvement and are indexed against 100 for data confidentiality reasons)

The vital cause now redefined the problem: Reduce thickness variation (i.e. sigma index for 100 hours S₁₀₀, representing population). Regular sampling (18 stacks per hour) and the measurement of stack thickness were begun. An X bar control chart (variable T_a) (See Figure 2) was introduced to track the variation of the thickness throughout the project. In the pre-project stage, the index of sigma = 100. Each phase follows the seven steps of problem solving to reduce S₁₀₀ by 50 percent from the prephase value.

Phase 1
The preproject section of the control chart indicates that the process is operating at varying average values of T. Analysis of variances (ANOVA) was conducted and the results (See Figure 3) confirmed this observation:

ANOVA
Source of variation	SS	df	MS	F	P-value	F crit
Between groups	892	2	445.9	137.0	7.27E-15	3.4
Within groups	88	27	3.3
Total	980	29
Figure 3: The variation between groups (i.e. groups with different process averages) was clearly the largest contributor to the total variation and therefore tackled first.

 

Figure 2 - X bar control chart – average stack thickness T_a

Brainstorming produced a long list of possible causes for variations in T (Figure 4). An Ishikawa Diagram helped reduce these 26 causes to 11 main fishbones.

	Possible causes of Ta variations
S Number
1	Recipe variation
2	Oven temperature fluctuation
3	Lamination
4	Cooking time
5	Ingredient 1 moisture
6	Ingredient 2 storage deterioration
7	Ingredient mix preprocessing hold time
8	Ingredient 2 strength
9	Ingredient mixing time
10	Temperature
11	Steam turbulence
12	Ingredient mix temperature post preprocess
13	Varying pull during lamination
14	Ingredient 1 dry addition variation
15	Gauging
16	Density of sheet
17	Ingredient 1 - component 2 quality
18	Mix preprocessing mix
19	Pretemperature - Oil
20	Water quality
21	Underfeed/overfeed mass
22	Ambient humidity
23	Ingredient 1 - component 1 quality
24	Fuel fluctuation
25	Number of layers
26	Pretemperature of Ingredient 1
	Figure 4. An Ishikawa diagram helped reduce 26 causes to 11 main fish-bones (variables x1… x11).

Hourly plotting of the X bar control chart and simultaneous recording of the corresponding values of these 11 variables was begun to determine the vital causes of variations.

When the process was producing the standard thickness T_st, the corresponding values for the 11 inputs were identified as “draft standard set.” “Spikes” in T_a revealed no clear relationships to changes in x1… x11. Querying line operators about variations revealed that each one had a method of setting the process. “Producing a good product is an art!” was the paradigm. To develop an accepted standardized procedure of line setting was clearly a key to reducing variations.

A multiple regression equation was developed to establish the relationship between the output variable and the 11 input variables, and used in a very interesting way to develop a standardised process and the mindset to use it consistently; i.e., T_a = I + a1*x1 + a2*x2 + … a11*x11

Note: x1, x2, x3 are key agricultural ingredient properties for which variation within a range was inevitable. The process therefore demanded that x4 … x11 be adjusted to offset these variations and produce a standard T_a (=T_st).

Whenever T_a varied significantly, for the given set of x1 … x3 values, the equation was used to estimate how T_st could be attained with minimum variations from the draft standard set. Repeated discussions with the team gradually made them question the large variations in inputs being made. A standard setting procedure evolved: When T_a varied, the operator returned to the standard conditions, observed T_a, and then fine-tuned the settings from that base.

The variations began reducing and line setting became easier and quicker. Within four weeks S₁₀₀ had reduced by 67 percent from 100 to 33 (see Figure 2, phase 1).
ANOVA analysis now indicated that hour-to-hour variations (i.e., within groups) had now become the predominant component of S₁₀₀.

Phase 2
Targeted to reduce S₁₀₀ from 33 to 16. Brainstorming and prioritising the possible causes using the “Weighted Average Table” method helped pick four vital causes shown in figure 5.

S Number	Causes
1	Cooking temperatures varying
2	Ingredient 1 quality variation
3	Sheet thickness
4	Varying temperature of ingredient mix due to preprocess variations
Figure 5: The group went through the problem/root cause/countermeasure/check result cycle for each of the above causes. Two of them illustrate the process and result.

 

Cause 1: Temperature variations were traced to freezing of fuel in the pipelines (Countermeasure—heat tracing), faulty design of a combustion chamber leading to damage of the combustion tubes (Countermeasure—redesign the combustion chamber)

Cause 2: Batch-to-batch quality variation of a major agricultural ingredient.
The team isolated a measurable ingredient characteristic that could identify the ingredient characteristics and developed different recipes to neutralise the effect of ingredient quality on T_a.

Likewise the other two causes were studied and their effect on T eliminated or minimized. The result was evident. S₁₀₀ reduced from 33 to 18. (Refer to Figure 2, phase 2)

Phase 3
Line operators were trained to draw control charts on the shop floor live and react to spikes. Not surprisingly, much shorter cycle times of response to changes in T_a reduced S₁₀₀ gradually from 18 to 12. (Refer to figure 2, phase 3)

Phase 4 – Upstream causes
Having improved far beyond its preproject expectations, the team wondered, “How far can we go?” A fourth cycle of seven steps was begun with the problem defined, as reducing S_100, reduced from 12 to 6.

At this stage, the team was struggling to improve S_100, using the approach used so far. A new insight was provided when it was noticed that the p value in the multiple regression equation of one input variable, product weight x11, had now become by far the smallest. Clearly, x11 had now become the vital cause.

The problem statement was restated—reduce the variation (batch average of x11) of X11 by 50 percent from the current state. Assuming the indexed value of sigma of x11 over 100 hours S_x100 = 100, the target was to reduce it to 50.

An X bar control chart for X11 was developed (refer to figure 6)

Brainstorming for possible causes, ordering them through an Ishikawa diagram yielded the insight that the weight was essentially dependant upon the weight of product before the cooking process. The vital root cause of this variation was found to be nonstandard methods of weighing the ingredients before mixing. Presized measurement vessels for individual ingredient measurements quickly fool-proofed this process. This change accompanied by the regular questioning of spikes on the X11 control chart rapidly brought down Sx100 from 100 to 45 (Refer to Figure 6) and correspondingly the variation S_100, to 6 (refer to Figure 2, phase 4).

Figure 6: Control Chart X11

Checking the final result—breakage

During phase 4, the team begun reporting dramatic reductions in

1. Scrap (breakage)
2. Repeat oven setting
3. Repeat packing machine setting
4. Pack weights variations

Data verified this “Feel” - Variable costs (ingredients, fuel, and packing materials) of production had reduced by Rs 6.4 million per year (approximately $160,000 or 3.5 percent of variable manufacturing cost).

A quality initiative story was presented to senior management. Post-project sustenance of improvement is often more difficult than the improvement itself. Reinforcing practices to ensure continuous improvement is therefore crucial for any improvement effort. The process recommended by the author is outlined below and was instituted:

1. Nominate line team – Production manager, quality assurance manager, engineering manager will meet every day and review and kill the causes of any control chart spikes.

2. Factory manager – Will review the control chart weekly and confirm that the team is maintaining trends and killing deviations promptly and decisively.

3. Senior operations management – Will perform a monthly review of control charts, killing spikes, and documenting findings.

The slogan “Kill a spike a day to keep the doctor away” was coined to keep the focus on continuous improvement.

TQM ingrains a way of thinking and working—a quality mindset—more than individual project or process improvements.

It's therefore appropriate to conclude this story with what each group member learned by the project’s end to get a feel of the extent of this change. This is what the team members had to say:

1. Believe in data collection and usage to kill problems
2. A systematic approach is key to solving problems effectively
3. Teamwork
4. The better the process, the less control required
5. Structured approaches to problems work
6. Quality is a way of life
7. Killing problems and maintaining improvement are the keys
8. Quality reduces costs
9. “We” can improve

It should be noted that members didn’t mention the technical aspects of biscuit making or TQM tools; they felt that TQM changing their way of thinking and working was the main benefit.

About The Author